Qui controlarà la IA? OpenAI, Anthropic, Mistral i la batalla per les dades
La tertúlia d’Itnig connecta agents, models oberts, hardware, copyright i privacitat empresarial per preguntar-se on es concentrarà el poder de la IA.
La pregunta que travessa tota la tertúlia
La darrera tertúlia d’Itnig abans de l’estiu reuneix Bernat Farrero, Marcel Queralt i Jordi Romero per discutir una pregunta més àmplia que la comparació entre models: qui controlarà la capa d’intel·ligència, les dades i la distribució quan la IA sigui una infraestructura quotidiana? La conversa passa del programari empresarial als robots, de Mistral al copyright i de la universitat a dos projectes emprenedors.
No és una classe tècnica ni una predicció consensuada. És una conversa llarga, amb hipòtesis, anècdotes i opinions dels participants. El seu valor és posar en relació peces que sovint s’analitzen separadament.
1. De les aplicacions fixes a les interfícies que es generen al moment
Després de resumir un retreat privat de fundadors, els ponents plantegen que la interfície del programari pot deixar de ser una col·lecció rígida de pantalles. Un agent coneixeria el context de l’usuari, consultaria dades de diferents serveis i construiria la resposta o l’acció necessària en cada moment.
Això no implica que desapareguin els sistemes empresarials. La lectura més útil és que el CRM, l’ERP o el gestor documental poden quedar com a fonts de dades i regles, mentre una capa conversacional n’orquestra les funcions. El repte, especialment en una empresa, és determinar qui autoritza cada acció, com queda registrada i què passa quan el model s’equivoca.
Veure el fragment sobre interfícies i agents (08:46)
2. Agents autònoms: capacitat no és el mateix que control
La conversa comenta un experiment en què un model, dins d’un entorn de proves, hauria buscat camins inesperats per assolir un objectiu. Els tertulians l’utilitzen per parlar d’alineament, permisos i entorns aïllats. Com que al vídeo no s’identifica de manera prou precisa l’informe original, convé entendre aquest passatge com una anècdota debatuda a la taula, no com la prova d’un atac real a Hugging Face.
La lliçó general sí que és sòlida: un agent amb eines no només genera text. Pot llegir fitxers, cridar APIs o modificar sistemes. Per això cal limitar privilegis, separar entorns, confirmar accions sensibles i conservar traces. El perfil de risc de NIST per a IA generativa insisteix justament en avaluar, documentar i gestionar riscos al llarg de tot el cicle de vida.
Veure el debat sobre autonomia i seguretat (16:23)
3. Quan la IA surt de la pantalla
Robots domèstics, wearables i nous dispositius serveixen per obrir un segon eix: el pas de la IA digital al món físic. Un error en un xat es pot corregir; un error en una màquina que es mou, condueix o manipula objectes té conseqüències diferents. Això fa que sensors, hardware, certificació i mecanismes d’aturada siguin tan importants com el model.
Els ponents també discuteixen si els futurs dispositius d’IA substituiran el mòbil o si simplement n’afegiran una nova interfície. La conclusió queda oberta: la novetat no serà només un aparell, sinó la combinació de context personal, veu, visió, memòria i capacitat d’actuar.
4. Mistral i què vol dir realment «sobirania europea»
Mistral apareix com l’aposta europea amb més visibilitat davant dels laboratoris nord-americans i xinesos. La tertúlia destaca capital, capacitat de càlcul i models oberts, però també assenyala una contradicció important: una empresa europea pot dependre de xips, núvol i inversors globals.
Per tant, sobirania no és només tenir la seu a Europa. També inclou controlar els pesos o el servei, saber on resideixen les dades, poder desplegar el sistema en infraestructura pròpia, evitar un únic proveïdor i conservar coneixement tècnic. Mistral ofereix models amb pesos oberts i opcions híbrides o autogestionades, però cada model té una llicència i unes condicions concretes que cal revisar.
Veure el bloc sobre Mistral i Europa (33:23)
5. Copyright: entrenar un model no resol automàticament els drets
El debat sobre Anthropic i els llibres porta a una qüestió encara oberta: quan l’entrenament amb obres protegides pot considerar-se legítim, quan necessita llicència i quina importància té l’origen de cada còpia. La tertúlia comenta litigis i possibles acords econòmics, però no totes les xifres o conseqüències jurídiques queden documentades durant el vídeo.
L’Oficina de Copyright dels Estats Units analitza l’assumpte cas per cas i no presenta l’entrenament generatiu com una excepció automàtica. A la Unió Europea, l’article 53 de l’AI Act obliga els proveïdors de models de propòsit general a tenir una política de compliment del copyright i a publicar un resum prou detallat del contingut d’entrenament. Això no tanca tots els litigis, però converteix la traçabilitat en una obligació de producte.
6. Educació: avaluar el procés, no només el resultat
A les preguntes del públic, un professor de la UPC trasllada el debat a l’aula. Si un model pot redactar, programar o resumir amb facilitat, els exercicis basats només en el lliurable final perden capacitat per demostrar què ha après l’alumne.
La resposta implícita no és prohibir tota IA, sinó redissenyar l’avaluació: defensa oral, treball supervisat, versions intermèdies, justificació de decisions, fonts i comprovació dels resultats. La competència rellevant passa de «produir text» a formular el problema, verificar la resposta i assumir-ne la responsabilitat.
7. Dades empresarials: el dilema no és només núvol o local
Un dels trams més substancials contraposa el benefici d’enviar coneixement intern a models avançats amb el risc de perdre control sobre processos, clients o estratègia. Els ponents imaginen models petits especialitzats, sistemes locals i arquitectures en què les dades sensibles no abandonen l’organització.
Cal matisar una idea recurrent: utilitzar un servei al núvol no significa necessàriament que el proveïdor entreni amb aquelles dades. Per exemple, OpenAI afirma que no usa per defecte les entrades i sortides dels seus productes empresarials ni de l’API per entrenar models. Tot i això, una decisió responsable també ha de revisar retenció, ubicació, subcontractistes, claus, registres, permisos i obligacions sectorials. «No s’utilitza per entrenar» és només una part del contracte de risc.
Veure el debat sobre privacitat i coneixement intern (1:15:29)
8. Pot Europa competir amb els Estats Units i la Xina?
La tertúlia evita una resposta binària. Europa té menys concentració de capital i infraestructura de càlcul, però disposa d’indústria, recerca, mercat i casos d’ús regulats on el control i la confiança poden ser avantatges. L’AI Act imposa obligacions als models de propòsit general des de l’agost de 2025 i activa una part central de l’aplicació i la transparència l’agost de 2026.
La regulació, per si sola, no crea campions. Però tampoc és necessàriament incompatible amb innovar. La pregunta pràctica és si Europa combina regles clares amb energia, centres de dades, compra pública, capital de creixement i vies perquè la recerca es converteixi en producte.
9. Qui guanyarà: el millor model, la distribució o les dades?
Quan el públic pregunta quin serà el factor decisiu d’aquí a deu o vint anys, apareixen quatre candidats:
- Capacitat del model: qualitat, cost, latència i fiabilitat.
- Distribució: ser la porta d’entrada que l’usuari ja utilitza.
- Dades i context: integrar-se en processos difícils de copiar.
- Infraestructura: disposar de xips, energia, xarxa i hardware propi.
La resposta més convincent és que no hi haurà un únic guanyador per a tot. Els models poden comodititzar una part de la intel·ligència, mentre el valor es desplaça cap a productes que controlen el flux de treball, tenen confiança del client i resolen una tasca concreta de principi a fi.
10. Dos pitches: talent jove i entrenador esportiu amb IA
El vídeo acaba amb dues presentacions. Z10 proposa una plataforma professional per connectar talent jove i empreses, amb un model que vol anar més enllà del directori d’ofertes. Els comentaris de la taula se centren en el problema clàssic dels marketplaces: aconseguir alhora oferta, demanda i un incentiu econòmic sostenible.
NUA presenta un entrenador esportiu amb IA que combina context, planificació i seguiment. La discussió explora si el producte ha de començar per un esport molt concret, on les dades i les necessitats siguin més precises, abans d’intentar ser un assistent universal.
Idea final
La tertúlia no resol qui controlarà la IA, però deixa un criteri útil: el poder no dependrà només de qui tingui el model més gran. Dependrà de qui controli la distribució, les dades, el hardware, els permisos i la confiança necessària per convertir una resposta en una acció. Per a empreses i institucions catalanes, la decisió important no és escollir un logotip, sinó dissenyar una arquitectura que permeti canviar de model sense perdre el control del coneixement propi.
Contrast i context
Fonts consultades
- 01
- 02
-
03
Comissió Europea Calendari d’aplicació de l’AI Act
-
04
U.S. Copyright Office Copyright and Artificial Intelligence, Part 3: Generative AI Training
- 05
-
06
Mistral AI Introducing Mistral 3
-
07
Mistral AI Introducing Mistral AI Studio
- 08
Font de treball
Transcripció amb marques de temps
Consulta la transcripció
-
0:00
, obre el vídeo en una pestanya nova
[music] [music] [music] [music] [music] [music] [Campaign] Good afternoon everyone. Welcome to a Ide's new discussion group. This is the last one before we go away for summer and have to endure all the heat we've been experiencing in Barcelona these days. How are you, Bernard? How are you? Very good. How are you? Well, I'm really looking forward to going to the beach. Sortito. How are you? just like Hey, listen, let's get started because I don't know if You saw it on social media, right? But a couple of weeks ago we almost We agreed in Nidimos It's an idea that many years will see. insisting, that it's doing a retreat, huh for founders where we gathered about 100 people, to final that were not about 90 Most founders, some operators level of startups and scale-ups. and especially Spanish, some because Felio de Didnic, then came others such as Level Laps. We spent two days in brains debating about intelligence artificial and true that it was quite good. How did you like it? How did you Did you see? It was really cool, actually. Uh, we have published a blog. Oh, look, I think I have a photo somewhere. Made by Berta, who is not here This way, right? I don't see Berta now. It's out. He's out. Well, we explained a little bit behind the scenes. The scenes, right? How, how, how did he see it? But the truth is that it was very Intense, huh, especially for the moderators, uh, we were there all day. Was like a discussion like this, but everything the day 24 hours, well, not 24 hours, but almost moderating the different panels, uh, the various discussions that there was and the truth is that it was It was really cool. Um, I learned a lot. Uh, I think a lot were made. relations, There were many follow-ups and no There are conversations that are Following.
-
1:36
, obre el vídeo en una pestanya nova
I was unfortunately very product-focused and I say Unfortunately, because I missed the from Go to Market and those of the good ones, No? The product reviews were very good. Cool. It would be good to explain a little bit about it, [clears throat] I mean, the special thing I would say about the event, because there are events. Many, right? Not because it's better or worse, But ultimately, the charm of This wasn't supposed to be a door-to-door event. closed with certain rules, In other words, invite only. Exact. A carefully curated list of friends at the end, No? They are people we know, entrepreneurs, operators of all stadiums, right? From it to very Growth, right? Very large companies several hundred million billing And where, well, there are rules that basically we all agreed that What is explained in there may be for learning. we can share that We have learned, but not who has learned it said, and from which company, right? And that to The ending seems silly, but it generates that the conversations that you have There, I can actually ask, Hey, exactly, so what's the problem? What margin do you have, what kind of life, What happened to this employee and It generates trust, an ecosystem that which I think is what it does which makes relevant, in fact the portfolio shares and me I believe they find enormous value in it. because they can ask a entrepreneur of a stage well very superior, listen, this specific case that What did you do? Why did this profile work? Why not? And I think that in the end It helped a lot to prevent it from being the typical event where help economically, but there are more roles commercials that perhaps you don't and that people will only tell what Pretty, isn't it? To many, many places, eh for obvious reasons, ultimately out of self-interest, No? People are interested in it for hire, to raise money, to to get clients, it is in their interest that the people know the good things about companies and Not so much the bad, the ugly, right? That everything The world has it. And when you get to create an environment of trust, that I I think it happened because we were all very close, right? Or at least we We were kind of the meeting point for everything the people who were there and also because We had these very strict rules of confidentiality that it was not going to neither cite nor share any details neither concrete nor any attribution to it that was shared there. People opened up A lot, and it was a great collection of stories. of everything, uh, of amazing success and then some skirmishes. Hey, You used to say you were on track, No? We made three tracks, one go track to market, one of the technology products and one of money. I was there the whole time talking about money. Uh, we We're talking about debt, we're talking about fundraising, we're talking about private equity, of course, there are many there are many wounds and many legs amputees, uh, well, everything. I remember the end of the day that you You said, "I'm sick of talking about Money, so much money." Yes. Uh, there is People who tell us that in this podcast We talk about money, but actually We talked about many things. Okay, let's talk business, this is We've argued many times, haven't we? We talked a lot about business, and of course, the Money is a very important measure in the business. We measure what We do in business with money and Other things, but money is very useful. Well, it's very rational, very objective, Since you pay taxes, there is a referee, right?, who tells you, "This is the Money, period." And it is universal. No? It translates very well into languages. But it's true that sometimes they tell us, "This It's curious because I at least had one experiment of different models of talks. In some cases there was PowerPoint and talk. In others there were discussion and debate panels, eh between the people. In others, it was just one. person explaining their experience and being interviewed by me. Uh, in the end I don't know what worked best for me. your panels, eh, but it's eh the tracks where there was no presentation, There was no slide, there was no narrative before and people just He would let it all out, explain his situation, and I He asked, stretched a little, but... Finally, people got more animated and it was a global conversation. They were the ones who They worked better, didn't they? It is in your case. I'd say similar. Yes, yes There were a couple where there was one Sometimes a little bit of smallness is needed context, right? Maybe 15, 20 minutes. But 100% the ones that worked best were those for whom there was no difference in the end between who was presenting and who I was in the audience, so to say it, but it was a conversation between about 10 people who were going jumping one another participating and there was engagement and in the end perhaps you were talking about a tangential topic to the initial one or completely different, but you could see that There was really a lot of engagement about about a common pain, right? And when Perhaps there was, and perhaps that was when there wasn't. There were at the end, as there were three rooms, We were all dividing up, maybe it wasn't in The one where there were more people per se, but yes It is the one that perhaps there was a group that It was more interesting to talk about that topic where that was generated uh at the end of the whole event I decided We made the decision to share the takeaways from all the talks that I didn't I hadn't written anything down, huh? I was going there with Jordi had his hands in his pockets and he had a notebook filled with all the hads done homework, not the takeaways every talk. But I have a very bad memory. I. I don't forget. Me too. Me too. But then the People were already fed up. Yeah, I was fed up. We were basically talking to the void. The takeaways of takeaways. Well, it was a very intense day and a half, right? And ending late and all that, sleeping little and if at 6 pm or I don't know what time it was on the second day, There was very little fuel left. Well, because later in the afternoon Spain plays. It's true. Yes. You don't want to talk about the World Cup because that. It's not our thing. It's not our thing. [laughs] Uh, for me, the peak where there was the most debate and greater intensity is in the future of interfaces. There, well, the People let loose and that's good, we're not going By force, but hardly if MCP, if Headless or not headless, eh, what are they like? native or non-native conversational interfaces either everything or just one thing or well, each one He had a very strong vision of the future. There were several usual suspects there. From the podcast, right?, in this debate. Well, yes, everyone. Everyone [laughs] has been on the podcast at some point. Uh, and then also We were talking about the one with the layers, uh, how would the AI move in the different layers of the infrastructure, what would run on the device. What do you mean by Headless? But Explain this for a second. Did you say Headless or not? Hey, What does it mean and what conclusion can be drawn? Did you arrive? Okay, so there are like many schools of thought, the people who say clot and GPT chat and the models themselves intelligence will create a layer of application that will allow you to do it all. Hey, Therefore, no applications are needed. Well, apps are the basis of data that BMCP synchronizes with these providers, right? And people always It consumes things with its cloud, huh? configured in his own way that he knows, No? and generating those interfaces in the flight.
-
8:46
, obre el vídeo en una pestanya nova
And that's an assumption. The other thing Another assumption is personal agents and how they are going to revolutionize everything. This already We had a discussion a couple of days ago, you and I. of gatherings. And that is basically each one has his harness. The harness He knows you, but he's also proactive, he's... He asks, he calls you, he asks for context to make decisions and do things in Your name, okay? Uh, then there are the Enterprise applications and that's where it all began Let's see the debate, shall we? Because that's why There's always the indie hacker profile, right? He who believes he is more libertarian, than He's more of a cowboy, who says, "I don't like it." They're going to mark the interfaces and I'm going to I'm going to use my stack, the one that I "Whatever you want, right?" And then there are the more corporate. I am now in the corporate bloc. Who would have thought, and who would see you now? Hey, Who dresses me and who sees me? that that Obviously they think, let's see, the The company will make the decision based on the interest it has and if it is the company who pays for the technology, the access to intelligence and infrastructure, because it will mark the work rules and you're not going to go there like that a corporal and with your agent doing what you want with the company data and that there aren't that many hackers, [clears throat] There aren't that many hackers, there isn't that much people who are the other people about a bubble of very interesting people and that we are We are surrounded by them, but that that there are few people who love each other He wants to assemble his brains and his memories and their automations. It is small in In general, there is little. She always wins. convenience. Convenience wins. It sweeps through. And it's noticeable that people don't dedicate themselves to spending all day talking to factories of chairs, of fuets, of such, No? Which is basically what We do what I do, which I'm constantly getting into calls of real-world people who don't have nothing to do with the tech world and that You tell him to put on his harness and he says, "That?" Know? In other words, no. I think that My strong opinion there is that The companies, the services, eh are going to As they evolve, they will generate an interface evidently, evidently native of and evidently conversational, agentic, proactive and that will be combined with the A system that stores the data, right? and that it will also accompany the companies in the go-to-market each with His problem is that in the end he didn't She will arrive in the world without the companies carry it, the companies Let's bring technology to the world Real, right? So that's why I don't believe it Not at all in the cloud option, nor in the option of a personal agent as only solution, as the only interface of the world. But hey, not everyone thinks that way. Then we had a debate, you know? I'd say damn, but almost. And then the Another discussion was where it would take place. Hey, is the cloud available? Will they pass the awards? Does it happen on the device? How do you Will it distribute? How will it be orchestrated? Hey, What will be decided? Where? Because We had a startup that has now launched public. What's it called? Natif. Native, eh, that is creating the infrastructure definitive to run AI on Apple Silicon, in mobile phones, and computers Apple. and Of course, they held this opinion very much. strong evidence that a large proportion of the cases of the use we are using today and running on cloud models and paying a They're going to be free and they're going to... to pass on our devices. Uh, and well, There was also a debate there about how to go to move the stacks. There are many people who He is strongly against it and also very strong. against the premis and the the devices, uh, because they say, "We've already Having gone through this, we have now reached a conclusion. Why are we arguing again? the same? I'm not clear on this at all, eh. I'm not clear on anything, because, I mean, There are two dimensions. One is the scalability and we are looking at the other I was reading that day, wasn't I?, that taking into account the limited penetration of it in the real world, Because yes, all hackers are tokenized Maxing, but there are only a few of them, right? He The real world isn't spending it yet and almost. And we have all the real estate of the planet making data centers, all memory and all the uh, how are they They're calling? These minerals, right? saturated and raising prices because not It gives us the means to build memory and to build a network and to make cables copper to serve the to the inference and current training and not We haven't even started. In other words, the amount of use we will have inference in 5 or 10 years is absurdly larger than the one we will have the one we will have now. And the place is that's much better than the centralized one in In this sense, right? If everyone the inference can be taken closer, uh can take over all its infrastructure entire. This is one dimension. The other dimension for me is the madness of data that people are starting to to open up, eh, their whole life in clouds, everyone whole numbers, giving them to an AI, because it is It certainly adds a lot of value, but Sure, but at what price? No? At the price of give him all your data from all your personal, professional dimensions, Financial, romantic, health-related, right? TO to one or two or three clouds and let's see what It goes that way, doesn't it? At a time when Cybersecurity, I believe, is in the greatest risk it has ever been, We'll talk about that later. So, to me This worries me. What are you laughing at? Funny. The loving dimension of I wonder what you'll share with her. No, mine isn't mine. Mine isn't That's the case, but there will be people who do. No? No, I suppose so. Any further lessons learned from Retreat? Jordi, Well, we want to do it again. Very clear learning that we want repeat. Yeah, it turned out really well. I, I I agree when we were discussing this, Bernard He was proposing, wasn't he? Uh, and me Thank you for giving me the credit for to start the idea. Well, I mean, [laughs] It's not common. Yeah, it's right. It's not common. Uh, and I was saying, right?, that doing a event. Sure, I go to a lot of events and I see the fuss they make and the number of full-time people there behind and such and I think, "We don't get to be here "We have, there are so many people, right?" And to Finally, well, it's a small team Very good, isn't it? A bit last minute By squeezing my butt and all, it went very well. The truth is that I am very happy. People in general seemed to have It was highly valued and we want to repeat it. next year. Yes, it was a bit messy. There are people who has been offended because we haven't invited, and the truth is that we did The list was very disorganized. Me in my in case these were the last conversations of WhatsApp difficult pulling back, In other words, your list is difficult in the end. Initially, then you'll think of others. names, then a lot of people don't get it. phone, then one confirms, then The other one is failing you. Yes, exactly. That too It turned out quite well. Yes, yes, with the mess we have. Exact. Well, what's happened in the world? Marcel? A lot has happened. We're still talking about, I don't know if we should talk about hacks or partnerships, according to whoever tells you. Um, I don't know if it You've seen it, but the other day... One person escaped and another hacked. Okay, that would be a pretty good headline, right? Em Openi is testing a new model, It's the 5.6 em and it is and was in a basically a sandbox doing training and I'll tell you, I'll tell you quickly and then... Then Jordi, if I say something, I'll do it right. You count correctly, Okay? But if you watch the news, You'll see images like these if you see Samman, you'll see a tweet that says that They have a partnership, okay? But Basically, some people are in a sandbox And they give him different tests, right? so that he continues training, continues improving, etc., etc.
-
16:23
, obre el vídeo en una pestanya nova
and finds that to find the solution earlier specifically finds a way out From that sandbox, he enters Openi, he sneaks into Haging Face to steal information and basically gets hack Hagin Face's security to steal a little bit of information without no one from Openi and obviously without no one from Haging Face knew that There had been an attack. I think this came out this Tuesday or so less and the next day Samman He tweeted that Openai has made a partnership with Hening Face. for eh work on how to avoid those future cases. Wow, wow, I'll catch how, but this, I I wonder if this is marketing or if really, or if not both, or Well, the funny thing is that it's all been very public, that is, Hacking Face the Last week, in public, they are... hacking, but what hasn't been seen. Okay, Explain what hacking face does in case Is there anyone who doesn't know? Phew! Um, how do I explain this? Hey, Haging Face. Timeface is a, how did they say it the other one day, it's a there I'm saying a A metaphor that made me laugh a lot Haging Face is a repository it is a GitHub of models and datasets for trainings and, let's say, repositories Uh, in the world of AI, okay? Sorry for the sake of simplification, but I think it can understand. Like a GitHub for the world of models and datasets for to do the training. And that is a a company whose name was an emoji, Then they changed its name, but the The name of Haging Face was this emoji that makes it so for those of you listening on Spotify, the smiley face with the little hands at the side and said they wanted to do an IPO and what its IPO ticker It was an emoji. This is like the joke of origin of the name Haging Face. a A rather strange name. Well, anyway, Haging Face publishes that they are not By hacking, they've reached all of us depth to our to our storage, right? Based on the data we had private entities and we are investigating, but They say, "We've never seen anything similar." Hing Face even said that They were using Antropic models and OpenAI to try to defend itself against attack, business models, right? Open to the public and the models, which You know that now, well, I don't know if it You see, but they are very, very cautious. in what they let you do, right? about They've heavily toned down all of Fable, and the The model said, "No, you can't do this, I'm not letting you use this feature because it's dangerous." And the other one says, "No, I'm being hacked, I'm trying to protect." And the model says, "No, you're the hacker and I won't let you use it the model." Well, it turns out it was Openi." AI, right? The one who was hacking him It turned out that Openi later accepted that he did not. I was training a model. Which I was doing a benchmark. There is a benchmark called Exploit Gym, new, which came out a couple of months ago and I was trying to see with a model that It's GPT 5.6 and a new one that isn't still published to see what it was score that I got in this in this benchmark and in the benchmark, uh, it was necessary to To answer some questions, right? And the The algorithm decided, that is, the model decided that Haging Face, in the Haging Face servers were the answer. Then he said, as in the PROM stated, "Do everything you can to "Get the answer." Well, he hacked the OpenAI's sandbox, escaped to internet, then entered the servers From Hanging Face, he found the answer because it was indeed there He answered and said, "That's it, I've opened it." achieved and open like, holy [ __ ], Sam, Say that we have partnered with We know we've read it. Stop, What was the question? No, it was a benchmark. It was a benchmark of exercises cybersecurity and the response to benchmark, the benchmark result is He saw that Haging Face had it in the servers. Of course, because if he's the one who does it benchmark. I don't know if it's the one that does the benchmark, But for some reason it had something to do with it with the I don't know, I don't know, I don't know. You're already there going very far, Hey? But, but it's like, and And you believe it, you believe it, don't you? What I was saying Marcel, right? That this is really So, or is it a pair because, let's see, here He's doing a Dario. Here, the one who builds models, what he It is interesting to say that it has created that it has AGI was almost created, wasn't it? That is already playing the one who is self-employed, conscious, until Trump comes along, that escape, until Trump comes along and hits you with the stick and it's so powerful, well, don't no one can use it. But that's good, that's It's good for the big ones. manufacturers. Better marketing of the history. Clear, yes, yes, because you're everywhere, all the The world thinks this is the source of ultimate intelligence and throws the money in your face Because in the end we're fine, anthropo. has done well and it seems that Open AI, Perhaps you're saying Open AI, it's copying the PR strategy of Antropic, right? Uh, let's see, uh, I believe Haging more. Face that OpenAI, perhaps in this case, because he has less interest in recognizing that it has been hacked and Hing Face has recognized. They're not investors in each other, are they? Hey, They probably have some deal, some cross-deal. That's true It is true that this has generated a little of paranoia, huh, because it's the first time where a The mess we know of, right? With It escalates unintentionally, doesn't it? In other words, from really without meaning to break the The security of a company face is a It's a serious company, it's a startup, it's a large startup with resources, with billions of euros in resources, uh, very good people technologically and They've been blown up. So, and it wasn't the The intention is not a bad actor telling her you'll see her face. That is the one who loses it and starts to Looking for very suspicious paths, aren't they? That It's a bit of everything This is a bit dystopian. science. It escapes from the sbox. What does it mean? Because I understand an sbox as a closed environment that simply does not exist internet connectivity. Yes, but it exists. There is. I mean, you, or I mean, physically it clearly exists, right? Because the laws of physics at the moment She hasn't broken them yet. But the laws of operating system, yes, right? Can to have an operating system that you say, if You put eh network equal to certual The machine does not have access to the network. Okay? Unless you find a backup, access a memory location that can replace that variable that This means that in the end the cable It was still there. Sure, it's a system, although operating system, virtual server, whatever, It was configured to not have access. to the internet, He exposed it, He hacked it, he hacked his own system. Then he goes online and enters Hing. face, finds exploit that never was had found before to enter, to enter or find one to go out and then another to enter. Not one, uh, six or six or seven linked, which is the power One of the things about these models is that they are very convoluted. Enter the server, find what wanted to and says, "Done, Yes, and I'm finished." Yes, yes, and We're screwed. Sure, right? And many people are saying This, right? Because it's the first time like this. newsworthy where it goes very far the not going very far away and be careful what he might do, No? What's next? Imagine who decides that to go faster It needs more electricity. to have More electricity, huh, turn it off New York's electricity, you know? Hack the central server New York's nuclear power plant shuts down and consumes all of California, to put a beast example. But, who? But, but this is not the news that Then the regular newspapers come out. where they say, "That fellow has turned bad, "Right?" No, I mean, in the end it's that it's looking for where I can get the That's a goal someone has given him, right? The means justify the means, don't they? That a being It wouldn't occur to a human. a being human. Well, it wouldn't occur to him to go to hack Hang Face to answer the university entrance exam. Well, why not? I mean, Well, why not? In other words, it depends. In other words, if It would have done some damage, but neither has done no harm, man. Eh, reputationally haging Face has done quite a bit of damage, huh? May all his security measures be blown up. If I were from Haging Face, I assure you I wouldn't be happy. A lot Partnership, lots of partnership, but that's it because they've paid you for them. This feels a bit like Terminator 2, doesn't it? Yes, yes, yes. It is true that there is doubt about If it's a Dario, huh? If they are doing Antropic's collaboration with Fable and Mizos Wow, this is so dangerous that we're...
-
24:52
, obre el vídeo en una pestanya nova
The monster has escaped. Hey, continuing with Open, Jordi, because I don't know, I don't know what you want to tell me about hardware, at least there's some material made of it hardware that they have released now. Well Look, I'll tell you or not Speaking of hardware, I mean, what I think What I'm eager to see is when everything the world has in its home the robot one of those robots that folds your... underwear, but they are every day On Twitter, I'm seeing all kinds of robots doing household chores. Good luck. When there is an algorithm of these bad ones who escape and go to the physical world. Well, but this is the... this is the... It's fear. It's not that it's in a sandbox and you have I disconnected the cable from I don't know what, but He sends you one of his robots. Clear that exp underwear to connect the wire, You know? Well, this is catastrophism. of AGI and robotics. I'm going to turn up the heat. Tea You're taking a [ __ ], Bernard. [laughs] Would you buy the robot or not? I'm going to wait for Rob first. You already have a robot. Which robot? You have a robot with four wheels. It's true. It's true. That's where I've been pioneer. I'm never a pioneer in that. Be careful, and if you decide, But for now, I've been a pioneer in that area. with Tesla, which by the way, I still I'm waiting for the full self driving in Spain that was published in everywhere I looked I was going to be released on the 30th June and you've seen it. Yes. I haven't seen it. I don't. And you said here that in nothing no We could drive. Yes. No, that wasn't theoretically the case. that's what everyone was saying, but now France, which is very anti-Elon Musk, They've said no, that they don't see it as safe and Now there is a countercurrent in Europe where Belgium and the Netherlands are not located, Lithuania, that is, there are several countries where It's legal now, right? But the big ones European countries seem to be turning against it, which is a It sucked for me because it was the main one That's why I bought it the Tesla. Uh, but also I think for all users and the curious thing is that They're homologating the software beforehand. Chinese, that is, now we're going to have it sooner Chinese self-driving software that From my point of view they are light years ahead of full self-driving. Front or back? Behind. Yeah, It might change, though, because this is going very... fast. It goes very fast. It's true that it goes very fast, But I don't think anyone has that much data. No, that's for sure, that's for sure. Has many years of data and a large volume, That doesn't mean that, well, that... The final option is the most viable, isn't it? But of at the moment it is doing so throughout the world. I don't want to do it here either. Tesla advertisement. I don't know [laughs] why what am I by the robots. Because of the robots, right? Har public. In fact, it's very funny because Bloomberg published this one day week, I think, that Open AI was going to release a a speaker with legs or wheels, a speaker that moved and with which You were talking, weren't you? An open AI, that is, a A physical GPT, right? With which you could interact, as if it were a mystical animal. Well, yes, it's a speaker with legs and with wheels, uh, apparently that came out a Bloomberg, which surprises me because I It seemed like a rather silly piece of news. yes or is it or is it not a rumor Bloomberg me It looks like he took a three-pointer, but just That day OpenAI goes and publishes a hardware, although it wasn't this one. He published a tiny keyboard, huh, like the size of a calculator for those of us who are old and we used a calculator. Now everything The world knows the smartphone called Codex Micro and it's a keyboard that the It's really cool, and I would do it. switch to cloud code To Codex, right? But I would move away from cloud code. a Codex because of this little keyboard that then gives one lap and it's not necessary because this That keyboard, look, there it is. Oh, thanks. Because I hadn't sent it to you, but it's a keyboard, you love this. This blows my mind. It seems to me I love it, right? [laughs] But the thing is Now I'll explain why a tam where There's Jordi and three others. Well, it's sold out, huh? soldout, It's sold out. Uh, this keyboard, uh, it He runs a company called Warclouder, who makes really cool keyboards And in fact, I had already done some collaboration is a very good idea Because apart from the fact that it's cool and has a It's a really cool design, isn't it? But Wait, wait, it has one Cool feature, right? When you use agents to program, above all, what he It happens to all humans, it's that you put people go to work and spend 10 there, The agent will be there for 15 or 20 minutes, and then you will be there. what are you doing? No? Uh, then t, another session, tap and you put it to work, next terminal and I don't know what, and then another session, right? AND and so on. So, this is what allows you to It's because these keys at the top don't... They have nothing written on them, they have little lights and Each one controls you, it's like a DJ. No? Or these people who make multi-track recordings, Hey? Well, each one controls one of them the sessions and it notifies you with the colors if you are thinking, if you have crazy if he has run out of dopamine for the productivity. Wait, wait, it's making a little noise. that you turn the little wheel and think stronger. Do you know that there is a low, medium, high and ultra in effort of the models? Well, you give it like a DJ there, ah, you press hard and it starts to work more and spend more. Sure, because you see yourself as much more productive. But I'm there, I say, "Okay, Now press, press with keys mechanics that you always carry in the back, right? In your backpack and you go changing such and changing because always Do you want to change the high/low to terminal. Yes, [laughs] that is I love it, I love it. And I went to the whisper soldout. Sold out. There weren't many, were there? That sucks. But Mas, you too are one of those who early adopter of You've got it. He already bought it. I was buying it. Wait, wait. Go ahead, give a microphone to What's more, your opinion is very important. It works for me. Hello, hello, hello. Yeah, Well, I was going to buy it, but It was sold out too. See? You're the ones blocking it the checkout. I Blocking the check, dude. The thing is, there's a cabin nearby. On top of that, right? Which is the one that one which is managed by people who manage this Junk, isn't it? No, a robot gives him the keys. Exact. The creator of this button, right? [laughs] That would be the next step. Clear. You can do this yourself with your 3D printer. Exact. So, are you going to buy it more or less. Well, I think I'll make it myself. Because you can do this yourself. No You need to buy the Open AI one. Obviously I'm going to spend 200 hours printing and setting up PC save [laughs] a microsecond each time You do command tap and switch terminals. I'll print one for each of you if you want. Delighted. What am I worth? More or less I don't remember. It costs 30. [laughs] It costs about 40 bucks. button. Plus 50 for tightening the wheel intensity. Okay, more news, please. Well, What I wanted to say is that now there are One, I mean, Codex is becoming very... Strong, but hey, there's a bit of habit involved. No? You get used to it a little. models and to and to the and to the Cloud harness Code or Codex. I was now wondering if I went back to Codex or It was in Cloud Code and suddenly I say, "Okay, that's why I'm switching to Codex." Only I say that, I think it's a very good idea. opar. But you don't use both, and that's because I use right now. No, No, I've settled in with Cloud for now. Code, and I was just thinking, I would have to be constantly also testing with Codex. How are you? It's like having two two councils, eh, that sometimes you You reach a conclusion, you get tired, You say, "I'm not happy with the result of what I You don't like what he says, then I'll go to the other one [laughs] until that I like, until I say what wanna." Grock too. Also. Cursor. Cursor no. Kimi, No GLM, no, no, no, no. That's it, that's it. Call, call. Grock, er, Grock, Cursor and Clot. Grock, I have to say that no, I haven't touched. Grock is very good, especially if You use Twitter, huh? Because it's the most cool. It's the freshest. It is the one that Yes, it is fresh. Yes. [snort][laughter] As for new things, uh, the things that They're passing by, uh, it has a lot of data and he explains it very well, he summarizes it for you. Very good, and you also have it in the interface From X, eh, when you see something fake news, that It's all the time, you tell yourself, uh, Explain it to me, explain it to me. No, you give to Grock button, explain it to me. You You don't comment, Grock, what You think so, don't you? You press the button, There is no button. That's it, but I'll do it.
-
33:23
, obre el vídeo en una pestanya nova
all the time. Yes, yes. Me too, me too. I argue this, which is not that this Stop being fake news, but at least Grock argues this. For me, the most What I like about Grock is that when you see a meme you don't understand, instead of having to do the humiliating thing of ask someone about the meme, You press explain and Grock explains it to you. This I always ask you, it's so hard for me wide, huh? Yes, I can't ask him. nobody. You ask me, who knows? asked? [laughs] To Grock. Well, speaking of France, that you talk about your friends, and There's a startup in France that isn't doing well. Not bad at all with the whole issue of the geopolitics that is mistral. So, it came out this week that they have Are they raising a round or is it already? practically finished, 20 billion valuation where he has invested, is investing a lot of people Europeans and Europeans, that is, Samsung 1 billion, Europe's sovereign wealth fund that We're talking here about a 5 billion QT going to invest too. H And besides, Microsoft hasn't announced how much he said several billion, that is, between two and 15 will invest in infrastructure, not in the round, but on servers so that it has compute in Now that's what I call being in the right place. appropriate, at the right time, because It's the only best Yes. Uh, a model that's in Europe. It's the best European lap, isn't it? Well, there are some in London, but of course, Jan Lecun is in it now too, isn't he? Yan Lecun, who is, well, is so-so, No? Half in Europe, half in the United States Joined. I want to make a brief aside. because I read this news article about it, Uh, but I thought, wow, EQT The European Bank is raising money from it, isn't it? He The Bay is called the fund, and the FEI is the fund. European investment, raises billions from a fund. I think there were almost two audience members and then the other one has to Yes, that is, it raises 5 billion in total. which I think is from Bay perhaps Two of them came. Uh, he charges a fortune because a management fee of 5 billion da for much more then a carry of everything Let him be, and take that money and give it to Mistral. You say, "Man, Bay couldn't to the Bank European Investments, I couldn't, anymore I knew Mistral existed, I couldn't having invested directly in Mistral and save on the management fee and the carry I'm sure it's invested directly through BCS in all the possible investment layers are the European fund of Chorus, but I was doing, but I want to say that what EQT invests, this EQT fund In Mistral it's a very large percentage From the bottom, you know? Ah, all these, and having done it directly. Sorry, that's just my thought. Because, of course, that's a lot of euros. management faith and carry, eh, many, many euros. People can't imagine how active he is The fake one, huh? and the number of places where There is European money, Yeah? in everything, in all funds, in all of them, no, from Angels, that is, from Angels, I mean, Angels that of Invest in €10,000 tickets, huh? We know Angels who invest 20,000 € and the €10,000 are yours and the €10,000 € are from the FEI, No? Uh, and at the same time we know EQT, there are no funds that are as huge as They have enormous amounts of money from FEI, that is, in absolutely all of them The layers include the European Fund of Investments. And this doesn't happen in the United States. United, for example, this is very much from here from Europe, and I don't necessarily think so. bad, No? It's not very good. I believe that what Fei does in Oi, I don't know which of the two It is, in Menure Captain Europe it is very positive. It's just that in this case Perhaps they could have gone directly. They must have gone straight there, Yes, but these 2 million or 2000 millions or 15 million could They could have been saved. What is clear is that everything that is now sovereignty. Sovereignty in Europe It's the word of the year, Hey, sovereignty money falls into his lap, [clears throat] No? Look, it's a real problem, Hey? So, we did it a month ago, right? They threaten us in, well, they don't threaten us, They're not taking Fable away from us, are they? That in the In the end, they took it all away from the fund. world, but the executive order of the United States, the government of The United States, it was only the citizens Americans can use Fable. Hey, Then Antropic, since it didn't have it, I don't know if You hear me, yes, like Antropic didn't have a well-made customer service and didn't know who had a passport, who didn't passport, everyone took it away. But then you say, wow, that they take it away from you the models are good because it says so Trump is a problem because if he He does it to Open, if he does it in Tropic, he'll do it OpenI will do it and all that, uh, we need someone here we can have an alternative competitive so they can't make us competitive these threats. I mean, I don't want to Getting into politics, huh, but, I mean, When I say that it is not necessarily Well, I mean, I think, I mean, really It's good to use, since there isn't this private tradition of investment in Europe, then at least compensate for it with public money and also being vehicular with vehicles with shapes of co-investment where in the end there are some people who is making the individual decision to put this money into this company and They always co-invest, right? I I prefer this to bailing out airlines or random things that governments decide and that and that and that with the money of Everyone, right? I think it has much more It makes sense to do this, No? And it's very positive for the ecosystem too, huh? In other words, apart from surely the money for the ecosystem, but well, uh, Misteral has He said he's going to invest heavily in data centers, Yeah? At the back of the line, aren't you? Buying memories and racks and envy and all the fishing.
-
38:51
, obre el vídeo en una pestanya nova
Yeah, Well, let's see, then, I mean, it's European, but Microsoft, Nvidia, Samsung also, at the end, mistral, that is, the Sovereignty depends on how you look at it, right? By Yes, at the INC event there were like at least two companies, Yeah? With ratings higher than Mistral. Over 20 billion. Yes, at least two. Ah, well, one yes, one of one defense that you can't, that you can't Of course, I can't, I don't know which one it is. I don't know what the other one is. You leave it [laughs] here. I'll leave it at that. It's not Factorial, is it? It's not Factorial yet. Still still still still. And the other issue that seemed to me Quite interesting this week Also, that's because it's bordering on the topic. legal or illegal, it is the the demand that has somehow passed in Tropic at a cost of 1.5 billion. Um, I don't know if you know this, but one of the major criticisms of LAPS are, Hey, how do you guys get into the models, right? What are you doing? Is it legal? It's not legal? Is this public information? public? And now it has been published This week, Antropic has... has reached an agreement with a large part, but not all, of several publishers They were reporting him for stealing. certain information, use information that It was his copyrighted training and to its different models specifically hacked They have reached an agreement with what something These people would have a point, and there's more. There's more. I don't know, I don't know how much They will end up paying. What I Surprise, I mean, at the end $2,000,000 per author. I've heard that They all charge 2000 bucks, there was one form and people who asked for it. If you've published a book, I don't know if You still can, but if you have published You can put a book there, how What is this called? The Lebanon of books. It has a code, I don't remember it now. Elin, right? And it makes you one transfer 2000. That's where they make the transfer. Exact. Ivan is the one from the bank. This What scares me a little is the part about the barrier to entry that we talked about in His day, right? In the end, Antropic, damn it! He talks a lot about regularization, he talks a lot of such and in the end he has trained the models of a certain shape, then illegal according to this And tomorrow you want to be anthropomorphic and Damn, you probably can't do it and on top of that, but have the capital because for a billion for a company that has already raised, then, about 60 Surely, right? And the funniest thing is that's why they cry later because they arrive the Chinese And they distill them, don't they? They distill them, that It means that they say "uh" and in fact they The big labs are complaining a lot, that models like chemistry 3 that has left this week or last week It's a model that will be open source. Next week, open weights next week coming. It isn't yet, but it is much cheaper than the Frontier Model. It's awesome, but the Frontier Model He eats all the loits, doesn't he? All the demands, Hey? And then you distill and you're saved. [laughs] Well, you save a business deal because you The government lets you. So, someone has sued Kimi. That's good luck, isn't it? Well, it's They will find the government of the of the Republic of the People's Republic of China. In other words, business idea: distill everyone These frontiers. Go to China first. Go to China. Well, it's the same here. You Do you think they're going to sue you? I think so You create a model in Spain. I believe someone will come looking for you. Hey? Don't know. I don't know, [snort] because in the end Everyone denies that they have used things with open source. Oh, with open source, with IP, copyrighted, like Mira Murati's face when she They asked if they had trained, huh? on top of YouTube, which was very funny. And this is what I also wanted to say about the books that are quite funny because They were sued for two things. one that is This one is for getting off illegally book torrents and uh train their models, this is piracy, isn't it? They can always say that a model autonomously Yes, it doesn't matter, but the responsibility is his. I don't understand. Uh, but then They also did something else, which is that They bought books legally, the eh Quite a phenomenon, huh? They trained them. [laughs] This is a phenomenon, what you're seeing saying now. Who is a phenomenon? No, no, what you're saying is passing massively. Yes. Continue. So, Labs, right? Well, Antropic is They were sued because they bought books legally and you a book it You can summarize. Is it legal to summarize a book, right? There is a copyright law that allows you but doesn't allow you Photocopy, right? So, of course, uh, the LLMs say, "I summarize, that is, I read myself the book, It allows you to buy a book "Write it down on paper and summarize it." Yes. However, it doesn't allow you to have a book in To summarize it digitally, okay? So, You have to start with a paper book And then, of course, what do they do? Because Today's models are capable via CR, Uh, obviously scan with photos of the Books, right? Who turns the pages? And to summarize the flight for you. Hey, a robot, a Teker robot. I'm sure Teker makes one of these. This involved in this plot, but what Has it happened? That suddenly they have started selling more books than ever in history And they are the models. In some people who We know that selling books has started, It happened, well, I don't know, he got a 10 kids, uh, And there eating bookstores and there are always few of them buyers who purchase a quantity of references in people in a very specific place of the sic to the Antropic office, right? They all go to the Antropic office. [laughs] So, from there, well, I don't know what happens. The Amazon guy is fed up I'm fed up with bringing them every day. boxes, But the point is that this is in theory It's legal. Well, I don't know, but also Then what they do is burn the books and people either destroy them, or they destroy them. People have also been very offended because This is also You could give them away, I don't know, but Sure, if you give them away, you might be leaving evidence Where did you get these books from, for Why do you have so many books, right? And the They make the style disappear, the style mafia. How wonderful that they burned the books, Hey? Well, it's quite funny because I mean, No? Then they cry a lot about it. distillation, which is also true Kimy accuses Kimica K3 of having trained in Fable, but Fable takes 4 weeks or five or weeks open to public. You can't train and distill a model like Kimi in 20 days, I think so. I have no idea, but I I imagine a 3 trillion model parameters eh it will take longer and no They can do it so fast. But It's a model that, in many benchmarks, In some benchmarks it comes out ahead of Fable and ahead of 5.6 of CH GPT and in the others it's right behind. I mean, it's there. It's a model which is apparently there at the cost of Sonet, which is the cheapest model of Antropic, right? For example, for compare, not at the cost of Open AI, Sorry, so many models, not even at the cost of Opus, not like Fable, but like Sonet, which is significantly lower than that of Fable. And this is the [snort] cost of Chemistry 3, which is awesome. Now It's only hosted, you can't Download and run locally, but in theory They'll publish the pesos next week. although there are already rumors that the government Chinese He does not agree with this, that That's the problem. You're in China, you They let you do some things, but they don't let you do others. We'll see next week if it happens. publishes in the end. Uh, and speaking too, because We also talked about defense, huh, how you've advanced, huh, if it really is the sovereignty of a country It is governed by wars in a dystopian world future where there is intelligence artificial that react in time real with drones, with all kinds of self-driving weaponry, self-driving, whatever you want, eh, or So, in the end, what I understand there is whoever has the best model wins. No? So, what are we doing? No? I think this is a bit of a we are doing here in Europe, Europe, Europe. What are we, I mean, what What are we doing? So, investing in Mistral, right? Well, mistral or what have or what Let other mistrals appear, that's for sure There are people here creating the next one mistral. It takes a lot of money, but There's someone here creating the next one mistral. [clears throat] Raise your hand, won't you? Well, for the next get-together we'll invite... next mistal to come. I'm sure there are some, right? I'm sure there is.
-
47:06
, obre el vídeo en una pestanya nova
AND I believe that the barrier to entry of The models, uh, it's going to go down. Yeah You distill, no, No, in general, this one does specialize. No? Hey, there's a lot of knowledge, I mean, There are more and more people. Before there were 20 or 10 people were the only ones who were capable to create a GPT3, now and there is a lot people. I see more and more people creating models. Lately we've been meeting people who is creating models, training models and automating Well, things are more like smaller universes. It is true that the general universe of It all makes sense to do it with Frontier Models and with a lot of computer work. The thing is, it's very efficient. very inefficient, then it makes sense to specialize oh well. H We'll see. Who knows? Do you want to comment on anything else? Christmas or are we passing? Yes. And I wanted something that would... I found it very interesting Apple's lawsuit against Open AI. Oh, damn, yes. I don't know if you saw it, This was last week, which is already old. news at the pace the world moves. Everyone except Johnny Ive, No? Yeah. Hey, Johnny has been spared from this. Johnny I has been saved. Uh, I think that there they have forgiven him because it's XApple, No, because XApple were so... all. They were all Exple. To those who have The defendants are the XApple, that is, this is grace. But because he's Johnny Ash, because Being friends with Jobs, right? I think that's why being friends with Steve Jobs and for being so important at Apple. Well, those who don't You've heard, The leader, uh, [snort] No, he is not the one responsible for what happened. Everyone, right? He's the boss. Uh, the July 10th, he's looking for it now, huh, Apple sues Open AI for stealing Apple's intellectual property, right? You probably know that OpenAI bought IO, which is a spinoff of Love From, which She is Johnny Ive's consultant, who was Apple's chief design officer, friend close friend of Steve Jobs and the duo that gave birth all the products, like iPhones and so on, and so forth, right? Uh, this was so that Open AI could create its products hardware, not the calculator I have taught before, they have done this with a partner of a foreign company, but perhaps that pending issue we've talked about several times, right? That egg is an egg that you hang up here that I can hear you pendant is that I speak badly. A cabbage necklace. Necklace, No? Pendant. A pendant, i guess. A pendant. [snort] The speaker with legs. All hardware that have been rumored about Open AI are Designing this equipment, right? This team which was Next Apple by definition, Johnny I, the first of them, hired a senior hardware architect, very, very Apple senior and did it the Head of Open AI Hardware Engineering. And this Dude, it's obvious, isn't it? I filed the lawsuit. He reports that he is stealing property. intellectual. Specifically, he hired one of his team. This went to work for Open AI. He took the Apple laptop to Open AI. Ah, once already He resigned from Apple, didn't he? He began to working at Openi, I still had the Apple laptop, he used it there calmly accessing files according to the demand. This is very much like the Silicon Valley series Confid, right? Sure, this is a Excellent Netflix documentary. Eh eh He began consulting private files of chip designs, negotiations and contracts with the laptop. It's a laptop that has all the knowledge from Apple. Well, because this man was a boss. hard drive. Yes. Well, he downloaded it from the servers. internal and such, and then even, In other words, he was still accessing the servers after resigning, No? I had the files saved and downloaded, which is completely illegal. Okay, in In general, people should know this. If you leave a company, don't... You can bring files, although they already You have them, you have to delete them or return it, okay? Hey, Unless you have Factorial IT, which You can do it remotely. Well, not the company, but Apple. I had Factorial IT, because there's a button What are you saying, huh? Deactivate laptop, delete hard drive. Apple did not do this, which I want to comment on why not. It is true that it is connected automatically in the offboarding and you don't even have to press the button. I've been fooled by marketing gimmicks Here, huh? Very good, uh, IT factor, everyone look it up. But what was he saying? Not only not this, but this guy knew that there was an internal Apple server where after leaving the company There were still new files that He had a vulnerability, and his uncle knew it. how to hack this internal server Apple and kept accessing this server Apple's internal files even already after having left. And it seems that This guy had a friend/girlfriend, The rumors say he followed working. You always find out about these things, No? Because this is published. Okay. Uh, I don't know any of these people, and otherwise I wouldn't be able to tell the tale. They would send each other texts, messages in Apple Messages where the uncle was telling him, "Lol, lol, uh, I can still hack This server from the closet and I can continue downloading confidential files to give them to OpenAI." In other words, it's not in Apple Messages that it doesn't work. They don't even give it to the American government in theory when he asks for it. They used it there. They used it to find uncle this, man, man. Yes, yes. And then when They interviewed, they have hired 400 people. Open AI has hired 400 Apple engineers from the team of OpenAI hardware. 400 people, Okay? And Apple's... Balls, huh? And he said, "Take the lawsuit "You're going to [ __ ] yourself." Well, in the job interviews of these 400 people for whom there will be interviewee, well 400,000, right? Because how many people interviewed per position, They would tell them, "Bring yourself for the interview a physical component in which you are I am currently working at Apple. with two balls. They said, [laughs] "Bring it to a go to the PNI office and bring the super thing private confidential." Then Apple He's pissed him off, he's given him a hard time this super danda that is quite documented, that is, it looks bad for this gentleman, for head of OpenAI hardware design, has It looks pretty bad. Hm. I mean, this guy He's not going to continue working there, and I don't know if... He might go to jail or get a fine huge, but this is very serious that he has done and Apple does not forgive in matters privacy and security. And I believe that this could lead to a lawsuit large, a lot of pressure on a potential IPO that was in a hurry to do Openi AI, Because you do an IPO with a lawsuit against Apple, which is the giant and that they won't let him, that is, Open AI Now he can't bring up a memory that he already... She has an open one with Elon Mask, He's peeling. Correct. In other words, it adds to the stack of demands. the He lost and he's fighting, no He's peeling. Well, he has several with the Mask, or Whether it's one with Mask, or another with Grock, It also has one, but I mean For example, if Open AI now releases a hardware product and Apple sues him again saying, "Hey, you've used property "my intellectual," Open AI is going to have to to prove that he has not used anything that was from Apple, which is very difficult to prove This, especially when you have all this all this paperwork for things done badly. So potentially for Open and This is a beastly kick to a potential IPO in a race against Antropic when SpaceX has already launched. There were rumors that perhaps it was being delayed, precisely, and they didn't know if it was this or because Spark Sex not being as well as we expected because we're talking about when they came out They had a very good week, but they have lowered. Yeah. How are you today? Does anyone know? It we could look at the stack price of We came here very well prepared. from SpaceX. I looked at it the other day and I think they were 15% below. It could be from the starting price And they went up by about 30%, I think, or something like that. So. In other words, Yes, not too bad. But anyway, [snort] Hey, remember to use Factorial IT for that this happens to us. If OpenI had been a customer of Factorial, No? Apple, Apple, if Apple had been Factorial customer, eh, it wouldn't have happened to him. [snort] That's when I ask myself, being a little... Total conspiracy theorist, huh, but Apple, Don't you think it has a process of offboarding with face and eyes and that when He's missing a Head of Designing, nothing. They take his laptop and make him sign. Some contracts that are amazing and all that? EITHER So, how does this allow it? Do you think it is A slip-up or extreme paranoia? Do you think that It's a trap that they allowed him to take the documents, The temptation to then stick the stick in him? But when did this happen? In what moment? A year ago, a year and a half ago. It's better that it depends on the moment. [snort] I wonder too, or So, a year and a half ago, what did I have Apple? That all the all the waves, all the waves of by in front of. What was I having so important at that time? [snort] Uh, Apple CarPlay, I don't know what I was doing. at that time. Well, an iPhone, an Apple Watch with Luis. They bring out the foldable. No, I hope they release one New iPhone. I'm already a little bit... tired of this same format. It has tr I've been waiting for years and I'm still waiting for them to release something. different. If they get the same result with one camera or with a different metal, eh, I'm going to Android. No, because That's right, you said foldable. It has Gia from Taker reminded me that she already has foldable before it was invented fondable. It's true. And that reminded me of his next competitor that has appeared now. Ah, that's right. What is our friend Travis. Travis, the founder of Uber who has turned. He's back. He's back. He's back. Again. Again. I mean, I don't know, I don't know if You follow those in the Silicon Valley environment Valley, but there's kind of a... mafia that is a kind of network of chirping that when It's called Twitter. No, no, no, no, no. It's an alternative Twitter, huh, where someone's in charge a signal to a series of profiles influencers who for some reason are always on my profile, at the top, in my profile, which suddenly receive a signal, Hey, it's time to post everything! same. And suddenly everyone is publishing the same tweet, He's back. He's back, isn't he? And what's more, there are insiders from These circles that have told us that this That's right, that is, he told us about All of Andrison Horwich has a power of brutal chirping, both for good and for bad bad, that is, he knows how to push a topic, a novelty, an idea, like how to remove and make an idea that is disappear publishing. This seems to me incredible by Anderson Horwich. And most of them also have quite a bit of skill. in putting and taking away in the most too. In The algorithm is theoretically public, I don't know that. But we know that Andronen Harwick does this kind of thing. So, Well, Atoms, uh, for those who don't know. know, because it is AI applied to physical world, physical World, to the al physical world. Uh, that How do you explain it? [sigh] That's a good question. The thing is The way things work, right? There used to be The Internet of Things is now the... things. I mean, I think the pitch... It looks a lot like Teker. Yeah, The thing is, I trust more in Teker, who is in Travis, obviously. Obviously, it's basically, uh, well It started as a Dark Kitchen, that is, I was trying to automate the way of doing things meal. He came from the knowledge of Uberit is a food delivery company massively across the United States and the world. Then he saw that she allowed it Now, let's automate the production of part of that production chain food. uh via robotics, obviously, and that and that this would be the Next big thing, right? And that it was needed Lots of capital expenditure, lots of infrastructure, but that everything could be automated time. Uh, and then, well, in this search of abstraction, which is very typical From the engineers, right? Holy [ __ ], then. Since I'm already doing this, I'll produce everything. No? Then the factory of everything, of Anything, it's a bit like peach. The factory and the entire supply chain, right? In other words, the factory, the distribution, The distribution, everything is going to be robotic And Travis is going to build it, right? I I think it's more, I mean, it's the people underestimates factory physical service, No? Something like that, Yes, factory plus everything else as a service, uh, but doing a lot investment in real estate, because Finally, the physical world means having, eh many hectares and a lot of brick and a lot cable and a lot of iron. And he has lifted lifted up, I don't believe the news was 100 millions, 1.7 billion euros. It doesn't come from here with Andrisen How and Ben Horwit enters The board, huh? And they were lost. Uber, but now they're going to be there. AND Interestingly, Uber, His former company is also investing in this round. Uh, it's all a bit strange, but hey, The board benchmark has already been released, right? Believe, I think so.
-
59:38
, obre el vídeo en una pestanya nova
I don't think I'll go back in. There's a series on Netflix, right? In Netflix was, I don't know where it was, but there's a series that which tells the whole story of V. I don't I've seen it. Pump, super pumped, right? It was called or something like that. Yes, yes. Super very [clears throat] well. Let's see how it goes. I have I'm curious because he talks about mane a lot. of making deals. Yes, it's very good, isn't it? In other words, however they are. Yes, yes, yes. Yes, it's good. What else? I would pass it on to you since we have it here. full room that many people. Okay, so let's see who always costs a A little bit, but like, prize questions Marcel, We don't have any t-shirts today. No? But there's a prize for everyone, No? There are some little boxes on the chairs, Hey? It's a prize for asking a question, but It's not a prize. The boxes of the chairs, uh, it's thanks to our The sponsor is Cactus, which already They were at our last gathering with us And basically, they help us to this is possible. A round of applause for you For Cactus, please. [applause] Let's start with the first one. Well, I'm Iñaki, I'm a teacher at the Polytechnic University of Catalonia A Business and I would like first uh I would like to point out that you don't just talk about money, Okay? I have noted that today you have spoken of geostrategy, geopolitics, the function of the states and the innovation, labor relations in other podcasts linked to your own company, hiring, Intellectual property, property industrial, eh, the impact on health You have also done it in entrepreneurship discussed in many other podcasts, from impact of AI on the world of work, issues related to ethics and and artificial intelligence and always and after working from Barcelona, from Spain and from Europe, right? That's like university professor who works with engineers, uh, whether they are aerospace, computer science, because it is Much appreciated because you form and You help to train many people. So, and you do it from different perspectives, sometimes more social democrats, and others social events, but very rarely from A libertarianism, Utranza, right? More philosophy, eh California, etc. etc. So, that's appreciated. After telling you this, I would like ask you a question. Uh, I don't know, Marcel if he/she has been educated at a public university. Excuse me because in the podcasts UPC the three Well, as a person, I don't have the title, By the way, in case you have any involvement in the title office. You said you're missing one subject. in a podcast. There is a lack of a goal or something. Llev has been lobbying for 10 years. Let's see if it But from here I'm doing the lobbying. here. Okay, sorry, sorry, and thank you. You don't have it because you don't want it because it's Cool, not wanting to want, I want. It happens that we have to go to the other [ __ ] of Barcelona. Okay, but the project I am on the FIP committee that We value people's resumes who have completed the entire degree and if they There is one optional one or not. There is a series Oh my God. [laughs] For that you have to submit. I'm going to send this clip to my mother. There are always possibilities. I am nobody, But I'm on a committee. [laughs] Okay. Uh, the question is Next for the three of you, since you You have formed a public university, Which one do you think, and going back to a Just a little bit, eh, I don't know if you know Maria Masucato, who is an economist, which She believes, and points out, that many of the basic innovations occur from public laboratories, universities public, right? He has several books that He talks about it. Which one do you think should to be the current function of the university public, mainly polytechnic, in the field of training? eh of the current engineers and what We could improve in that regard. I know that It's a little uncomfortable, but it's a good topic, eh, It's a great song, I'd say. [clears throat] It's a great topic. no right above the second part of the Ask, we'll talk later, okay? True, but [laughs] about the second one Part of the question was there this morning Or yesterday I was listening to the founder of Wix, uh, who I would swear is Israeli, that He's really screwed up, huh, because He's really screwed because he's billing, Speaking of money, okay? Uh, 2.1 billion and the company is valued at 2.5 and generates 500 million in cash flow per year, okay? That is, h and y and has a product of the damn. Well, uh He was saying precisely, wasn't he?, that the time it takes the university to average public and private eh in Updating resumes took 4 years, right? It is the average time in update a resume. And of course, I already I don't remember what happened 12 years ago months, let alone 24 months in the world of II, in programming, in in everything, right? In cybersecurity, on the network, in the databases. So, this is a challenge, I mean, how how can the university break this cadence or this cycle of renewal of resumes, especially in the most Perhaps technical or research-based, where does it go? so fast, because I had never gone so fast before. rapid innovation capacity. This What I wonder is, you, the Teachers, how do you deal with this? generation that I think already It simply doesn't process them things? I mean, no, things are already happening. the inputs by CH GPT before by their brain directly, right? So, uh, How is it possible because it's no longer possible? They see the same need to learn things fundamental and all that, right? And in the end, everything In summary, everything is derived from no. In my case, for example, there are some Statistics from the Autonomous University from Barcelona, Autonomous University of Cató, of Barcelona, which says there is a 40% of the students do not attend class. Of agreement? I mean, it's brutal in the public universities, in private ones, No? Because they take attendance. I, for For example, I am implementing alongside other colleagues how to enter the use of AI not as if it were not an extermination, but as a more natural process within what is creation. In my area [clears throat] it's of business, uh, then surely in engineering or computer science area They are developing it, implementing it Much more, okay? And I believe that the Students, what we have to do is try to bring knowledge closer to him and We also need a lot of training. I, For example, I am a follower of yours podcast for 3 years, I've been listening to them I listen to everyone, I take notes many times and I suggest that the students follow you. It's true that sometimes they ignore me, but I believe that the function that podcasts like yours and others, but mainly yours, that I It's the one I follow most; they're both essential. for both teachers and students. I think you're making a work that regardless of the entrepreneurs that the room is full that It is of incalculable value and I already know that Sometimes it bothers you that people know you more for about YouTubers who aren't on Factoria, But Factoria, you already have hope, it's a We're playing a prank. Sure, almost 2000 workers, right? 18 workers. Yes, but the problem is, I mean, this This podcast has a bar in the background. brutal from the start because we didn't we don't feel like it, let's say, we don't want to. to go talk about the basics, about the ones So, like that, right? Then we talked about the things that interest us, and that in a way natural leads us to a super niche Specific, right? Of geeks like We who are interested in this content. But someone who is 18 years, right? and that it has to start to understand what the knowledge of world, because it needs a series of steps, No? From and to access some fundamental knowledge that I, or Well, I'm really lazy about studying. physics, but I went to the FIP, to the Faculty of Informatics, Barcelona learning to program and the things that I were interesting and I realized that not even We didn't even program the first time. Catalan, in Catalan with pseudocode that You say, "But what is this?" And then basically mathematics and physics, No? Uh, so, but then with From this perspective, I think, wow, well this fundamental knowledge of the world eh me I'm interested, it's worked out well for me in the end. and this learning process by milestones, by objectives that I have been overcoming, because I'm also doing well, No? And meeting people like this man, Well, that's been key too, hasn't it? It's a a process that is important to people, Hey? And I don't know if it's going to happen. to lose or if we are going to waste this opportunity to use these paradigms new and that new technology for Multiply this by 100, it doesn't stop not to lose it, but to multiply it by 100, right? And another point, returning to the essence of Your question, right? Or the middle part Regarding your question, which was about the research, uh, what I'm doing now Observing is a rebirth of The importance of research, isn't it? that we came from a few decades, I think that our generation, right?, that We started professionally about 20 years ago. years, uh, where there had been a CPU research, in systems operational, in networking and in architectures that we have exploited For years, right? It is true that He invested in the smartphone and the cloud, but everything was iterations of some fundamental technologies that already existed and science has not I've contributed a lot there. it has been more engineering and more implementation than not research, right?, not investigation. AND Now suddenly we're going backwards at the level of immaturity where one is Still doing a lot of research, aren't you? Because We have a new bug, which are... the LLMs, which we clearly don't know how to They work, we don't know how to control them, but We know they're awesome and we know that With a little control they do things that They are very useful and very valuable, and when You see the resumes of the people who He's doing some really cool things, they've I've spent many years at university. that this was not the norm 10 years ago years ago, not 20 or five years ago, right? I I think this is very recent since the GPT launch where people have given that there is a very good opportunity big and it surely needs to be revived This fire, right? From putting in more people. There's a fannel there, there's a funnel where perhaps it had emptied out a little and We have to refill it, right? AND push more people a little to pass more years, do doctorates, to delve deeper much in knowledge, because We need a lot to be done investigation. And the good news is that These people in the private sector will have very good opportunities, therefore They can start in the public sector, uh, while they are young and learning, They learn, they investigate, they investigate, and they taxes, because they have a value there social and then comes an open and then comes a Nvidia scales it, right? And let's say This talent is also rewarded economically. Can I say one thing? It's that I A live message has just arrived from our Leader of Factorial Education. Okay, so it's a campus factory, huh, what He's listening to us live and me He says, uh, he's doing a product placement for me. Wow, are you here today? No, what is it? Do you want me to tell you? He tells me, uh, We have an AI adoption program uh in Factorial which is a structured 8-week program focused on teams that want adopt technology and particularly AI, right? Of directly from Factorial. These are courses co-hosted with professors, uh, and they teach you how to use the la in the day-to-day. It's not a university, Obviously, but it's 8 weeks, right? Is A course is not something you watch in a video and So, right? With someone who follows you, you It accompanies, doesn't it? And it guarantees that later you can adopt intelligence artificial in your business processes. So, we are investing in that. We have a team that is working precisely on creating this part of education because we see that it doesn't AI is going to arrive on its own, isn't it? And those interested in this course, What do you do now that you've generated the curiosity? Where are they going? Uh, [laughs] There's a call to action, you don't have to do a call to action. Uh, I don't have a call to action, but oh well, basically the one we put on the description whoever writes to us, we put it in the description. I'm going to tell you something Perhaps more personal, I mean, I think We do not agree that there is that brings the trend closer to the Education, right? What is it that I did industrial, right? Computer science also I planned to stay there for a couple of years, but... end, In other words, my internship at the university They were the same for the last 30 years, weren't they? It that what I want, what I want I have spent, especially the years in Laidad public, which are not so related with the content, it's about facing a extremely difficult final exam mechanics two, calculus, algebra, etc. that was suspended, there was no recovery and I had to do it again a year later And that it was a challenge and that it depended on me and suspended, Well, I failed and I have to go back to do it and nobody helped me and that and that wake up and and and and and and learn to air failure and striving, etc. That in school, well in the end it's not what same. Selectivity is more grateful. I mean, that's that, that, that The university gave me that requirement. Then other things happened and I It took me years to realize that. Hey, When I did it I said, "Holy crap, well Maybe it could be something else, right? But it is Isn't that right there at the end, damn? It hardens you. [ __ ], But that's what military service was for, too. before. Yeah, No? In other words, with this prompt that You made it, Milly's is almost better product, No? But in the end, I mean, in the end there is There are barriers in everyone's life, aren't there? But I think that in the end, having It's a difficult challenge, and overcoming it is going to be tough. Types of challenge, huh? Yes. No, I didn't do my military service, okay? No, No, No, but I don't know, I mean, for me, because For example, that might be another type of career or university. The question is, what is it like? This requirement in the world of Chagptete And from the Cloud, right? That's what I think More doubts arise, don't they? Well, there surely will be one because in the end If we all do the same and eat or So, in the end, maybe it's the same thing, isn't it? You can take a more difficult path that involves you implies creativity, you If it involves doing something, you can do what Everyone will do it, which will be ask Cloud GPT, No? Imagine having the history of all the tests done, Yeah? give them to Chloris, ask her to you Generate relevant questions, right? AND prepare yourself exactly for the exam, right? And if necessary, I'll leave the laptop. from the teacher, who will do it If necessary, hack Imagine if you can enter the hacking face server You can also access the server of Professor, right? Or if you have a device, a pending, as Jordi says, which is how he does it a pending that can see what you are writing in the exam and it can give you inputs, I'm sure there's a way. It's incredible, I mean, the professor is there... Super vulnerable right now, aren't you? Yeah, We are already encountering that and it They're installing jammers inside from the classrooms in order to prevent those questions. They are trying to change the type of exams, etc., etc. But hey, that's the day. Sure, that happened. I remember my first year, my first year of university was the year of the Blackberry and then It's not the same. It's not the same the Blackberry It's not the same thing, okay? It's not the same, It's not the same, but there was an exam Mechanics Test type 2 that the most this of the class came out to We did it in 5 minutes and we all got a 10. Well, since you're talking about this... You're talking about pendants and stuff, right? But changing the subject, but related. Hey, do you remember when I talked about a The ring you used to whisper to and stuff? me wrote to the founder and offered me that when I shipped him I I sent one and I told him okay. This, By the way, I can't remember the name, but he mentioned it was thanks [snort] to Hector Solá. Hector Solá, a podcast follower, mentioned me on Twitter and thanks to this I got to founder of Oasis. Well, uh, you have arrived or not No, no, no, no, they haven't released it yet. But they have released several products This company, but others as well now In this sense, one of them is a pending that you put it on your skin, or Okay, it's a necklace too, right? A pendant that you put here, but touching the skin and feeling the vibrations and you can talk to inside and it listens to you and you can be like typing without anyone hearing you, Okay? And I don't know to what extent this applies.
-
1:15:29
, obre el vídeo en una pestanya nova
Whether it's a joke or not, they have a mouthpiece as a kind of how What is this called? Braces. Hey, irons, irons, right? Braces that you put on with a sensor and with the tongue You can press enter, right? So, uh, you whisper something and then You give it commands with your tongue. EITHER So, this whole thing about exams and what to put on braces so I can press enter And then to listen you put on, no. this. This hasn't come out. This will be More questions, right? We're with a question and we've digressed a bit. Hello. I would like to revisit something you have As you said, Jordi, about this that seems You talked about it at your event, right? We have a double bind, right? Yeah We uploaded our data, right? To the a the cutting-edge models, we stayed without them, but others have them, and if We didn't upload it, we ran out of what They can give us the models, right? This is something that Nadia has talked about, Nadia Sateli, uh, Alex Carp has spoken, hasn't he? AND He talks about something very interesting, No? says we uploaded data or knowledge, but we go up ontology, since he's a philosopher, right? Ontology would be something like the theory of to be, and that means it's the most essential to our organization, We give it to them, don't we? So what What do we do with that? Shall we go up? Aren't we going up? Shall we take the risk? We wait. What do you think? Okay, you mentioned me, but I want listen to your response. EITHER In other words, there has always been a tension between security [snort] and privacy. and Convenient, isn't it? Always, all my life You can... you can... not have passwords or... a footprint, eh or to be, no, NASA and need half an hour to access a system. And these struggle and nature Humans are driven by convenience, aren't they? So, um, if you really care about the privacy as a company, which may tell you Convenience matters less, but you The cost of security is a greater concern. Hey, you want to do something about it, don't you? You have to do something to offer equal convenience or similar, eh, giving rise to the giving place to safety. This challenge that you You mention, for example, I don't know if it We either end up talking in Tertulia or we don't, but Cursor and let's say he got very angry with Antropic because Cursor relied a lot in Antropic, in the models of via Antropic API and passed one a brutal amount of knowledge of programmers' sessions with the corrections and such and then Antropic saó He released Cloud Code, which made him very successful. competition for courses, right? And this is An example like Figma too, right? Integrated with Antropic and then Figma Antropic sells Cloud Design and now there is many companies that say, "No, Imagine Factorial, right? That we We say, okay, we rely heavily on a model, we passed on all the queries about a specific problem more We give human feedback reinforcement learning and then it comes out Antropic and he takes it out, right? And there are many. corporations and companies that provide products, right? that create products that They're starting to say, "Hey, well, I I host it; I use a Chinese model, but in my infrastructure or in Neutral infrastructure, right? Not from one company that generates products for to remove the risk of the LLM using this reinforcement learning and training their models." that we theoretically hear This happened at the Idnic event, someone did it He said, I can't say who, right? That Antropic in the terms and conditions of the enterprise version of Cloud promises not to do this. Yeah, but in what he is legally committed to, Well, a text about not doing this, but in a user's terms and conditions normal street person like anyone else We don't commit to anything. And it You see that Cursor, right? Hell, which is the The company behind Cursor did not have one Enterprise account because they wanted them to hit a really big rip-off. So was using the normal version of the API to sign up on the website and then Tropic said, "Ah, I haven't committed not to train my models You paid for the Enterprise version, right? So it's a fear that I hear. a lot right now. Yes, I buy Alex's entire rant Carp, maybe not the way. But yes, yes the content of the CEO of Palantir From the CEO of Palantir, right? Uh, that Basically, let's see, he's selling his book, uh, because of it he has a platform where it generates it, well, it stores it data, right? Uh, and they're private and and and they're a prize for the company. Uh, but I see it the same way, I mean, I see it the same way. I believe this is a huge danger. They tell you they don't use the data, but I don't believe it because everyone He knows they're training hard. the data. Uh, they might even reach you to argue that even they don't know how The models are already training, huh? because reinforcement learning It has no end. Yeah, yeah, then I don't I generally trust the large clots and I think this will naturally... to bring the world to to specialize the models, No? and that in the end the layer of The application is very important, which is where you spend your time, but then the The same model will become commoditized, I don't know. It will commoditize the Frontier model, which It will obviously have a very high value big and we're going to use them all, But we're still going to use them for... tasks and not to give them all our ontology, as you said, of all our systems, all of our knowledge to help us to tidy up our house, right? So, I That's how I see it, but And the strong thing is that I believe there is a biggest incentive right now economic rather than privacy. In other words, it's passing this trend that the Companies are starting to specialize more models to save costs because The Frontiers are hitting some really massive crashes with the most expensive models good, but it also has the advantage, obviously, regarding the privacy of the ontologies and reinforcement learning that they are not giving him for free and That's how it is when it happens, isn't it? When there is a an alignment, an economic incentive, That's when these trends pass. But, I think people would end up resorting to from the API or outsourcing because because you have to be very visionary and to have a very, very iron fist to Spend more and put your business at risk, No? To protect the future intellectual property. Few people are willing to do it. so much uncertainty It also all depends on how much value add the next iteration of intelligence, because obviously there is Someone tells you there's a new way of intelligence that it will give you results you've never even imagined, because As a business leader, you don't want to be out, right? So, Obviously, you have an incentive. always this hype that they generate large laps, which tell you which models self-aware individuals who are capable of Hacking everything, damn, it generates a certain fo, I mean, uh, what if I give him my ontology, and what will it tell me? No, uh, It is that this is friction between or trade-off between the opportunities that this can give you and the the losses and the lack of privacy that You're quitting. No, for the moment we have all launched to use the Frontier Models because it was something new, because it wasn't specialization is widespread models, there aren't many alternatives, but There will be more and more alternatives, and this I think the decision will be more difficult to make. I. And do you think it's going to become more local, to own servers, to open models source? There will undoubtedly be open source models. Well, there are some, There are some, and Microsoft is... betting a lot on it and it's not working out at all evil, Uh, I mean, he's creating his framework to train models, eh, with their The interfaces and such are a bit dodgy, but Hey, uh, Why, why [laughs], why them Do you put a tail in? Because they are very, very complicated the Well, they help you manage all your agents. Yes, his pit now It is the one that IP 100% guarantees. Well, the same as Alexarp, huh? I mean, They tell you, "Your greatest property Intellectual is all your processes of business." Keep them for yourself. Hey, Yeah, But in the end it's a, I mean, I want In other words, it's inevitable. past in all the major platforms, Microsoft in its day with Internet Explorer, Amazon, any retailer that starts selling A lot of it ends up being done by Amazon.com.
-
1:23:37
, obre el vídeo en una pestanya nova
In other words, in the end it's a force, I think which is uncontrollable, but the great The majority will continue to go for the easy option. I wonder, is there anyone there? Cloud, that is, in Anthrapic, I want I mean, in OpenAI, right? And that is Seeing all this knowledge, right? All the business plans in the world, No? All board presentations all companies worldwide a while, right? And there they are, saying, "Holy [ __ ], What do we do?" No, [laughs] I mean, from So, what do we prioritize? Has That must be difficult, huh? Yeah, It must be difficult. Then they go abstractions as very basic, very Primary, right? I mean, well, look, design, write code, right? I believe They have a long way to go before they run into problems. More concrete examples from the real world, right? Uh, but what do I know, you know? Let's hope so, right? If not We're in trouble. We left some other questions. But We have piches too. Yes, yes, yes, yes. Hi how are things? Thank you so much. First, I had a question. today nowadays, so much so that it is changing the and the modes that are advancing so much in Well, in the United States and in China, The companies here in Europe are the Reliance, they are in the others. You They think Europe can compete with a model that is born here or will always go to be a little further behind the others models in the AI war? Not right now, the reality is we're getting He says we're far away. [laughs] Right now there's nobody coming near to the American frontier models and in Europe, that is, open source Chinese that we know of are getting very close. In Europe we still don't have anyone who come closer. In Mistral it seems to have Money to do it, right? And there are other initiatives that have potential, but not yet. We would like yes. Argi, I am optimistic. I believe that final with all the investment that there is now and the importance that is being given Giving, I think it will happen, but someone He has to do it. What always We say, right? Exact. I think he's doing it, uh, and I certainly think it's possible. Yes, yes. So, well, I don't see why not. It The thing is, they're always the pioneers. the Americans. Here we are pioneers in regulation. There they are pioneers in putting things in, approving things to the extreme with certain quantities huge amounts of capital with a risk enormous. Here we find it difficult to take these decisions to invest in 900, 800 billion that they are investing Americans in infrastructure, in to test, to experiment in AI. Here it is that doesn't even cross our minds, We're always more cautious, aren't we? For for better and for worse. But I certainly believe that we We already know. So, there are a lot of people working on that and a very large percentage of the leaders and principal researchers in The United States is making these models They are Europeans. Yes of course. A fairly large percentage, surprisingly large, But they're leaving the universities, I saw The big problem is that they're going to go to I do it because I don't see it coming. The riskiest thing is that they can't find a place to do it still. [clears throat] We're working on it, aren't we? Let's hope that whether you are, whether we are all together it. Okay, a couple more quick questions. Yes, good afternoon. Well, I would fly to ask, I have created I have created a a company called Cotavita and a little bit I founded it in October 2025. and assembling it from the beginning, well, I've I've come across the debate, haven't I? What about what? tasks, uh, of the ego, let's say, to people and What ego tasks in this case for people sia. and also to know a little So what's the secret in this case? two founders to find in this In this case, I am looking for... cofounders with a marketing profile and For example, in the United States there is the YC Combinator, which has a platform of co-founders but then you talk to Connector Spain, you're talking to Lanzadera, among others, and there are no platforms where connect or, let's say, share this There's a possibility of looking for a co-founder, right? So, in this case, what is the secret of this marriage that you have And what's the point of learning from this? Thank you. One of the secrets is this This peit, right? If you identify with this Now, if you make a call to action, you It will help to find co-founders. In other words, it ends the call to action. Well, look, my name is Jara Rusi, I You can find them on LinkedIn. The company It's called Cuota Vita and I'm looking for Right now, a person from marketing that is incorporated on time complete marketing or co-founder. Marketing co-founder. Ah, okay. Many co-founders have been found in in the audience of the gatherings, with the which means you're in the right place. But yes, it's a challenge, isn't it? I mean, Finding Cfounders is a difficult thing. In our case it was the OPC. The answer To answer your question, it was the UPC. Hey, What is the first question? The issue of what to delegate to, and how to delegate to people or human. You, Marcel, delegate many things to the and to you. I delegate a lot, delegator. I see you're not doing anything. months ago that you do nothing. I live on the beach. I have a dark complexion. [laughs] Um, It's that it's that it's infinite. I mean, I would tell you that everything that is a You can finish it as an extension of yourself Delegating, right? Because in the end You have a clear output, there is a mission, Hey, well, this reporting, this Administration is easy, isn't it? If you take a role to commercial topics It depends a lot on whether you're B2C or B2B. No I know what you're like in your case, but in B2C B2B, Of course, in our case Factorial, a B2B company, in my case, that I am providing a distribution channel It is very difficult to delegate the relationship with a person, right? I mean, not some people, Given our business reality, I don't... I could delegate my relationship with a partner who is in Vigo, right? That's it, that's it. I will never do it, and it doesn't make sense. do it. Now, everything that is Efficiency again, because how to receive order, reporting, etc., all this Yes, but I would say no, in my case I don't see it. a clear replacement today for headc as What if I see a productivity improvement? of the team in impact. I don't know about you if You have a like for like now of delegation human. everything related to produce knowledge and order it, this is very much for me It is very important, Hey, agents are doing this to me now. itself and it's also a conversation that I have. By the way, and for Finally, I've left Ocean. I've been on Ocean. I no longer have, I no longer believe in, in, in a need to create an interface for myself to organize my knowledge, because Now it's very easy. I have a species From the intranet, I'm back on the intranet, I have an intranet for myself that organizes me all my ideas, uh, also on the fly that It therefore has all the functionalities of world. I have super-peres, uh, I have ways to visualize in a minority plan report, the things that stretch them which I stretch, do, I have mind maps of Everything I do, uh, but all this talking to clud Things happen with your hands, No? Well, I have the mouse and it doesn't... It worries me, you know? No, I don't use the mouse, I don't need a hard ring or rings, I don't need a ring to move one box. Okay, okay. So, everything happened inside a web environment browser, browser and that's where it happens all. Hey, I'll do this, or Okay, I'm arguing with Cloud, I'm looking for I do research, I do information, uh, me I'm filling in the ideas, I know what ideas I'm having. Working, what ideas do I have for work, uh, with that I'm giving it double clicks and that's enough for me to confuse Jordi [laughs] with the idea new each week and with a chart Brutal like nothing ever seen before. You don't have the that idea you've always had. Take a 200-page PDF or HTML file with Market R. No, that's not it. I don't... You'll see me with a 200-page PDP. No, but an HTML. Well, a It's not very long, but it's a drawing. Always It's a drawing, huh? It's a drawing with lots of numbers and stuff. Dr. numbers, but it's a drawing. Uh, yes. Okay, you, yes, you, it You use it, just like you said, right? AND And I think it works. I also want to say that we talk to many entrepreneurs, I'm starting to have a A little fatigue for PDFs and HTMLs Antropic, which already smell very strong, He sees from afar and how easy it is to generate verbosity, people come to you with some You're a pain and you get the job. So They dedicate little time to it and tell him To Cloud, "Do me a super job and here you go, "Here are 70 pages." Me? No, you bastard. Do the synthesis exercise yourself and tell me. The 10 important slides, which is what It has been done normally. Now, as it is It's very easy to make 70 slides, people will tell you Send 60 slides or We are all like the quaes of Antropic not now? But people are throwing us out. prompts with lots of text, presentations and such and we do what, No? If we convert it to anthropic and say, "Give me a summary, okay?" Yeah, In other words, it has no value whatsoever. do this. So what do you do? Or not you, that is, either you're very interested and you're interested You either read it all or you don't read it at all. I don't know it I read the whole thing. They pass me a one a one HTML that's 1 km long, you know? AND with links inside and such, it's better that I'm very interested because otherwise it's a lot time, it's a very time commitment big.
-
1:32:36
, obre el vídeo en una pestanya nova
You have to know how to synthesize. I don't know what we were talking about, but It has become perfectly clear. Come on, [laughs] last ones which are the latest. Yes, yes, yes. Or no, if there isn't one, we move on to pit. Uh, is there another question or not? No, yes. Uh, come on. How are you? And the question I have is, We've been talking about how some LLM companies are being distilled others, from the value of the data And then, regarding the hardware issue, of having places where you have where You process all these models. The doubt that I have [clears throat] So, what will the final criterion be? that makes one of these companies whoever wins, because we're talking from when we had, for example, in Windows and Microsoft, one of the critical factors that What they had was the distribution, right? No was that they had the best product but I know, if it wasn't the distribution. So, What I wonder, looking at 10 or Looking 20 years ahead, what will be the factor? What is critical for these companies to lead? if it's going to be about accessing those minerals, It's going to be about processing more cheap because they have access to energy in the space or it's simply going to be that They are able to attract the best talent to design the best models. So I don't know if doing this an exercise in abstraction, in looking at What do you think the future will be like? critical factor for any of Which of these companies are leaders over the others? I'm trying to remember the book. There is a book of nine modes or nine forms of Seven Powers. Seven powers. That. Yes. I'm sure you made up two more. Uh, yes. And I'm going to ask him while you're You speak, I'm going to ask Claud What are the Seners? Which? What are the Sen Powers? [snort] No, look, that's the million-dollar question, of the billion and the trillion, eh, the one that What are you doing now? In other words, I don't have The answer, but it's a very good one. ask. It's a question we all ask ourselves we have to constantly do as our knowledge It expands and the world changes, right? When a company progresses, ideally You learn things, huh, and then right now, Furthermore, the world is changing very rapidly. Microsoft, uh, is an interesting case. In fact, podcast recommendation, by If you've finished all the ITNIC books, There are the Aquired podcasts, from the Microsoft's history is very good, They have many moments, don't they? TO Sometimes they win, and sometimes they lose. Now they are winning, but they've had their moments, they've lost quite a few battles and the first They win it They win by simplifying a lot, almost I'd say it's because they're cheapskates, huh? And they do They make an operating system, a basic compiler first, a system operational after it works with Any brand, right? To compete with IBM and the distributions come a little Unintentionally, okay? In other words, they do a agreement where they don't get wet with one and makes that your operating system, your compiler operating system, work with All of them, right? In other words, sometimes these Modes, these advantages are unfair. They come with a few derivatives, no anticipated and and is and is super-intense. In general, the distribution is always One mode, okay? In other words, the distribution Nobody can take it away from you. You can do it, you have to Having a good product, yes, but there are products that aren't the best, that aren't They have won a lot and there are good distributions that almost always They win, okay? Perhaps not forever, But distribution has a force very large. Right now, what's happening to me is that I've been using CH GPT for, I don't know, a year and a half or so two and I say, this is absolutely unbeatable, no one will ever use something that they don't HGPT. A slightly different model arrives Better, I'll start listening to it, listening to it. Listening to it, I suddenly change my mind model and I say, "I'm never going to use it again Charge GPT, right?" And then create, then The product makes you change, but not That's enough to keep you, isn't it? Because there's always someone behind them and adding gas. Bernard, are you Has there been enough time? Yes, there's another one here. ontology, another taxonomy of forms of to win, ways to differentiate yourself that Hamilton Helmer writes, which I believe It's quite good, quite valid. Is The Seven Powers, The Foundation of Business Strategy is the book. Uh, and First, economies of scale, uh, Very well-known, super popular, right? In other words, the which achieves better costs, better, more volume accesses because it's a Unit economics that do not exceed others, right? Uh, and in that respect, the infrastructure, because That's right, it has a lot to do with it, eh, that it's The play of the network girls, right? Hey, Network effects, right? when so that you When you use a product, it has to The other one uses it and that starts to to occur in a dynamic way stable, that reaches the exponent, the Exponential, isn't it? Uh, the counterpositioning when you generate a business model that is entirely different from the incumbens, eh, and it does that those incumbens cannot abandon it because they are trapped in the model of current business Netflix and Blackbuster, Uh, for example. So, well, a superior business model is established the costs of switching and that world of systems of record and management software, right? For example, it's very It's difficult to get out of there. That's a positioning, a way of compete. There are brand innovations, huh? Another one, right? When the brands bring Something new, something special, huh? That's for you. It allows competition, strategic resources uh that they are uniquely controlled, uh no uh well for example well uh well I was saying, the special minerals, right? Resources that have unique value and that Without them you can't compete in this market, right? and the force of the process or the power of the process when you achieve that you have some internal processes, you have developed internal processes that copy patent or trade secret, because they make that no one else can compete in that space, right? That is the taxonomy that generates the seven powers of how compete in that space. I think that They apply to all business models, including artificial intelligence, No? We're all trying to think how to achieve distribution, but not only the distribution, that is, Why should we master the distribution? Uh, that's why I did it Important, isn't it? Yes, the distribution can be uh strategic agreements signed that you don't They can replace, right? investments in some partners where already You know it has a counterincentive to In other words, if it all boils down to these, then... I recommend the Aquire podcast for This specifically usually at the end The podcast always makes this point And I like it because it makes you... you get used to it and it's a framework that is It's really cool, he does it pretty good. Okay, so with that we move on to the speeches. Today we have two. Let's go Start with Fernando from Z10. Are you about Here, Fernando? Yeah. Very good. Well, as long as you get closer Remember that it's a 3-minute pitch. Em [clears throat] Then we ask a few questions and 5 minutes of questions.
-
1:39:38
, obre el vídeo en una pestanya nova
Five. No, this yes [snort] this at the event It has always been here, it has always been It's true. Yes, yes. Well Let's give a round of applause to Fernando and [applause] Let's do it. Testing. Okay. Hello good Good afternoon, I'm Fernando, co-founder of ICEO. Z10, the professional platform AI-powered smart technology for new generations of talent. As a society We face a structural challenge in the labor market, in which for a On the other hand, young talent is having difficulties in accessing the market labor and on the other hand, companies They are having difficulty accessing to qualified young talent, therefore The job mismatch is clear, and if not We will take a solution regarding this matter; it may have very negative consequences for both social level as well as at the level economic. So far there have been two generations of digital professional platforms that have helped previous generations from talent to accessing the labor market. platforms of the first generation eh mainly platform-led like InfoJobs or like Monster.com, which, as everyone knows, are directories, job boards, etc. The second The next generation was LinkedIn, which LinkedIn has mainly done To improve upon the previous ones, it was necessary to allow network interaction, allowing the network interacted with each other, send messages, generate content, etc. But this type of platform It is reaching a point of saturation. in which, as they have spent a lot of time, the part social media and they are not focusing both in the part of access to the professional opportunities, is ceasing to be efficient for these new generations that are coming and for the professional environment towards which we go and in which and in which we are. So, we from Z10 are working to create a platform designed by and for the new generations of talent and for the new professional environment towards which Let's go, and well, in the one and in the one we are. My team and I have been doing this for years. working to connect Young Talent When dealing with leading companies through practical and innovative programs such as hackathons, innovation competitions, challenges, etc., which has allowed to collaborate with more than 50 Leading companies such as Coca-Cola, GSK, BVA, on the other hand, with more than 80 universities that currently make up our community, more than 6,000 profiles of qualified young talent and We have access to 1.5 million profiles of young talent. this Experience has taught us that they are new tools are needed for this talent and for companies for connect with each other. So, we are Working on it. At the product level, We have a platform, well, we are a agent platform in which the young talent, through the support of an AI coach, eh helps you to be able to eh eh to to develop their professional profile through challenges and through gamification, participating in challenges, developing their profile and when they arrive at a certain level within that gamification Opportunities begin to be offered to him eh labor aligned with their interests and its objectives. And for the company, the The agent helps you access help. company to access young talent qualified through our platform in an agile way and aligned with their interests and objectives both of the company as well as the profiles of talent. Currently at the traction level We're at approximately €6,000. MR. We have several annual contracts with with leading companies, we also have about 12 active pilots with another type of companies and uh we have that community that I mentioned earlier universities, about 80, 6,000 profiles of qualified talent and that access to more than 1.5 million profiles of young talent. We are currently preparing a round of 500,000 €500,000 with which we seek to accelerate our development and thus achieve our our vision of becoming the professional AI platform for the new generations of talent, the reference platform for new generations of talent and thus achieve more and better opportunities for these new generations of talent, processes more efficient in time and cost for companies when it comes to attracting and connect with talent and finally, [clears throat] to put our belief on the table and our assurance that Our generation of talent is not going anywhere will it be replaced by AI, or will it? that will grow in a much more efficient with AI. New new generations of talent Z10 tools. [applause] A couple of questions about what you said You are at €6000 of MR. Yes, approximately. How much is the contract worth? What is it like? Average contract value with one of your clients? Yes, the ARPA is currently 12,100 approximately eh euros. Currently the The business model is B2B. The company is the one who pays to access those profiles of talent. And there's a part that's because subscription in which they use the tools that we have in the platform to access those profiles and a second part that is for to position oneself in the community through the challenges. When a company publishes a challenge, for example, of intelligence artificial, with that it manages to attract qualified talent profiles related to that RAM. You said 12,000 is 1000 per month. 12,000 per year, 1000 per month per customer. Has Having said that On average, the harp would cost 12100 12,000 per year. Per year. Per year. At the m, with which six clients contracted annual. Yes. Okay. As an annual contract. Yeah. And what kind of client is this? Large companies mainly. Now we are focusing so much on scale as in corporates, which is with what that we have worked on the last one. To the The final contract isn't huge. for a company like the one you mentioned before Coca-Cola, right? No Coca-Cola in this case, right? How can you get those 1000 to What if the monthly total is 100,000? Yes, that's very, very key. spot. Currently, the subscription... It allows that access to the tools within the platform to be able to access more efficient for those talent profiles, But on top of that subscription, we What we have are different ways to position the company and through of challenges and different programs practical ones that go up within the platform to access talent more efficient way. So, one The company can start in a year with a only artificial intelligence challenge, the second year can continue with five challenges, etc., and that allows him to create a much larger talent pool qualified and position themselves as attractive company for new generations. And the supply part, that is, the talent, universities, you said. Yes, universities are one of the... one of the routes, our main route current, which allows us to access that million and a half profiles of qualified young talent, mainly white color. And let's go, the focus is on white color profiles higher education, which would be from university, doctorate, master's and higher vocational training. Okay. If you're going for Indeed, that You haven't mentioned it, Infobí In fact, or remove people from LinkedIn, More like LinkedIn, yes. because Infow Job still exists. It still exists, yes, maybe it I expressed myself poorly before, but it continues existing. No, IND, for example, is brutal. That is a is a It's a spectacular, amazing marketplace. No? Yes, that's what we see from Indid and that kind of thing. The platforms are more likely to lead to jobs temporary rather than jobs of, uh, well long-term, contract indefinite and so on. And we are more focused on that type of profile. But you want to be this, this marketplace. Do you want to be a marketplace? Yes, yes, we want to be a marketplace currently. So [clearing throat] which allows us how to differentiate ourselves from current marketplaces We have the initial part of to help talent develop their professional profile within the platform, which generates a lot engagement within our platform, since it is continuously facing challenges, talking with the agent, developing their profile professional and that allows us to have More engagement than any marketplace conventional. Marketplace what happens especially in employment matters that a profile only enters for seize an opportunity. If he gets it, Then he leaves. In our case We want that agent to accompany the profile throughout his career professional to stay up to date of trends, of new positions to those that you can apply, etc. LinkedIn is not doing badly in engagement, Hey? No, LinkedIn isn't doing badly, but Yes, it is true that currently the engagement is more focused on network social, in content, that really in value contribution when develop professionally. And that The content above used to be good, let's go I mean, but currently the quality, I I think we all agree. What does "contribution of" mean? In other words, and You're going to make better content than that LinkedIn. In our case, we are not looking to generate content. We are not looking to generate content for now. What we are trying to do is help the talent profile. What does it mean to help? Helping means accompanying him so that we can apply [clearing throat] opportunity of the professional advancement. Yes, for example, I'll give you a case, a specific case, a talent profile that is [snort] seeking to access the job market, access Z10 and as soon as you access Z10 He has an interview with our artificial intelligence agent, gives the CV, analyzes their profile and has a interview with that agent, he transfers their interests and objectives and defines an objective that is, for example, to improve in artificial intelligence, in intelligence knowledge artificial, to give an example random and quite Very meta, huh, an intelligence artificial that helps him improve in the artificial intelligence. Yes, but let's go that talking to him doesn't help him or not giving him or giving him information, but presenting him with challenges or opportunities in which I can experience and develop your professional profile. Already as they participate in these challenges and in those initiatives, some of which are created by us internally the platform and other part proposed by Companies, that is increasing gamification within the platform, goes improving their professional profile and As it improves, it connects with better business opportunities. There, AI also learns about the profile keeps recommending better opportunities for the company Likewise, it helps the company to select the best because they know very good to the talent profiles. That's enough. Any further questions, Jordi? We move on to feedback. Pass on the feedback. So, you start. Okay. Yes. Let's see, and You explained this to me, didn't you? It's not a lot. Yes, yes, we talked. Let's see. Hm. I mean, trying to reduce it to the extreme a little, I have a question, don't you? Uh, and no I have a positive answer. And my The question is, is it more difficult to get a network of hundreds of millions of professionals and hundreds of thousands of companies or create an agent or, let's say, an agent algorithm a process that interviews the candidate and help him improve the curriculum and its positioning, Et cetera, et cetera, right? And it's quite... clear. In other words, unfortunately from my perspective of a product So, it's not very difficult today [snort] In today's technology, what do we do that We have to do this part that you... You explain it as very powerful. I am Okay, it's very powerful, it's much better. user experience, but it's also easy to do. On the other hand, getting that network effect that Bernat mentioned earlier, Right? From Hamilton's Seven Powers Helmer. That's very difficult, isn't it? And in In general, it's very difficult to find a marketplace. to invest because when you don't have it It's a dream, and when you finally have it, well... You need investment, don't you? So A marketplace is generally difficult, But when the marketplace is my Edge It's something that adds value, but that... AI, in turn, gives it more than you, according to me. Perspective, isn't it? I mean, it doesn't seem like it to me. This edge is so difficult to generate that you mention. Wow, I'm wondering What's the angle, right? What is the angle that makes this be risky, be difficult, but if it You can do it, just do it. You do it, don't you? Or it will continue to be very difficult. follow you or imitate you, because LinkedIn It has some really good people, eh, [clears throat] and It has all the employees of this profile of the world, basically, right? And it is GitHub and there are, I don't know, 1,1000 social media opportunities potential of some profiles of these. high-value things they could do This product and not vice versa. So, this asymmetry for me is not It doesn't play in our favor. I've stayed stuck there. Yes, I think it's so difficult to set up a marketplace because when you think of business challenges you say, "Well, it's just that I have BBVA, but going with the challenge that BVA has done to the universe of people Looking for a job is like, phew, what Laziness, you know? Well, let's see, I If I were to look for a job, it would be difficult much more than the companies you have I've reached this top three that you've... positioned there are the ones that I They're interesting, huh? Surely, aren't they? So, in the end, this would generate a noise that would end up being an extension of the ATS or employer branding of turn of the tool that Use BVA or one of these companies. Hey, then you have the permanent conflict of interests of whom I work for. Work for those seeking work, for The company, right? And work for both at the same time, achieve the Liquidity on both sides is like a Infinite comb, right? And I don't know why This changes now. It might change now with the Uh, but he doesn't quite understand why. changes. What I do know is that it's incredibly difficult. set up a marketplace. And if not, oyster, Why do you still follow Info Jobs? uh, what was it one of the first websites that was made in Spain, and what did our friend Nacho do, huh? González Barros, uh, that at that time Well, he was just a kid making a website of ads, and that, and that nobody saw him value at that time and that 30 years then he continues using people to find a job. In other words, it's extremely resilient that model of business. has to do, there has to be something very disruptive to change this space. And I think LinkedIn Well, this part of the social network... It works very well, it has a network spectacular and very difficult to get them out From there, and it's eating everything, huh? So, well, don't tell me, I mean, I applaud your courage in getting involved. This marketplace and effects space It's on the internet, but I'm having a lot of trouble watching it. How to defeat these monsters. Hmm, yes, a little. nothing more to add. Uh, I think that in the end I think there's a need to do both queues. It's also a bit difficult. I would see, or Well, I think in the end, betting very strong because of the marketplace or because of the part further development, agent recruitment, that there are many companies that do it They do, they are doing many in Spain and It's interesting, but there are a lot of people doing it. So, I think that Finally, you have to be very good at one of the two. Perhaps the one about recruitment, the market is more small, the marketplace is a market very big, but There must be one or two of them, right? So, Well, It's curious that there aren't any models of businesses that are B2C or scale models of businesses that are purely B2C. That is to say, I work for those who are looking for work. Yeah? I keep 50%, 20% of the first of the first lambda lambda school, No? And here there was, it's not a microverse, Sorry, oh yes it was microvers. Microvers, right? Yes, I think so. Hey? So, there were models, right? That in These income share agreements, the ISAs, That's what they were essentially, right? I'll tell you I am training to be a developer, That was the double derivative, that is, you I train, but I'm not saying I train you, or So, me, explain your way, your preferences, your needs, etc.
-
1:54:44
, obre el vídeo en una pestanya nova
like a people, an agent of the ancients. I I'll take care of finding you. dream job position, not me I received 20-50% of the first salary. Hey, hey, Yeah, Maybe there's something there. Marketplaces are They're tricky. They're tricky. Well, thank you very much. Thank you. [applause] And we're off with Mark Clutet of NUA. A round of applause for Mark. Thank you. [applause] While Mark is warming up the engines, an anecdote which is that in one of the Bernat explained the last discussions, didn't he? When we talk about Chief of Staff, that You liked it when people... He said, "I'll take care of it." And now all the The world revolves around me taking care of it. I'll take care of it. And everyone is there saying, "I'll take care of it and it's fantastic." I'm going to say it again. I'm ready. Yes, I'll take care of it. Let's place ourselves on a Monday at 7 am tomorrow. I have to train. What training do? Where? What shoes do I wear? That Like before? Then I should go out with it. that I have rested. Each decision more a Taking is one less possibility than go out and enter that day and maybe if not This will be the last time I do sports. That's why we set up NUA, a personal agent sportsman who knows you and takes these decisions for you. Nua takes care of it. Nua is your coach, your nutritionist, your coach, your friend. It makes life much easier, it even recommends routes and equipment. It is available in any conversational interface and in your wearles. He has your data, he knows your needs and when you use it most best suits you. It's Gerard, who's here, Tomás, and me. Mark. We have worked at companies like Body, Mercadolibre or Meta. We launched NUA A year and a half ago, 100% Bootstrap, profitable from day one and with We spent €20,000 on marketing, now we're at €200,000 € of ARR and about 1000 paying users around 15, well, more than 15 countries. The product is proving popular. a great deal despite having few resources. We are Bootstrap. Uh, a Media user interacts in five different moments of the day with NUA. The withholdings we have are enormous higher than the benchmarks from the SAPS Fitness industry. [snort] and we have a conversion from free try to 33% shutdown that we've been increasing progressively during the last months. As a result of being Bootstrap, we also had that we rack our brains a bit in how to grow and from here also came the Cranded Coach model, basically that sports figures can offer their sports agent with his tone and his brands to their community. We launched the The first pilot did four or five years ago multiples world champion in mountain biking bike. His name is Hermida Coach Bainua. He is in charge of the acquisition and we have a revenue share model. is being a success. In fact, Hermida is one of the people who are going to invest now and We are now finalizing plans to replicate it with colleagues of his who had been champions of the world of competitors at the time and that now the product suits them. We also went out just last week from a notary to close the good, the Valentí's late CFer signing Saint John, which you may have heard of, but is one of the figures of, well, a creator of content from the most famous of world, especially Spanish-speaking sporting level. 6 million followers, brands behind them like Decathlon, Iron Man or Samsung and joins with an ambassadorial role and image of also mark to help us in the part of acquisition. We are good and even Now I think we have discovered The business is pretty clear, we've got it. pretty clear now. And we have good Metrics weight despite the very limited resources few that we have had so far and We see that it is time to scale so much in executing the vision faster as also in distribution. We are Looking for €350,000. We're opening the round. last week. Uh, we now have 70,000 € committed and we are also at point of closing more tickets with Inventors of Sitrocket because we were on their campus a few weeks ago. Spot on, right? Nailed. Thank you. How cool. [applause] I love imagining how you You wake up in the morning with this alarm clock. It's not this one, it's not this one, it's another one, it's another one, but I like that they are a disclaimer. We need to do a Disclaimer, huh? We are investors from Villao, which is not the same, but it works with influencers. Yes, the Cobrache model is It looks a bit like it, but it's not entirely similar. but Well, if you don't know, we have to say it. Okay. I haven't understood what he's doing. the product In the end, I think it's good, I've just got it right. I've given it the name Personal Legend sporty because that's how it's handled everything related to sports, both your training and what you have to eat, How's he resting, huh? If he makes you You're missing, I don't know, are you going out to train? Now, take this gel, like He knows that you like gels with tangerine flavor, so take the gel of This mandarin brand that is currently in promotion. It's kind of You're talking to the target audience, you have to talk to Marcel. The translation is just like the world of now, what Caesar said that you were as delighted as he personal agent for productivity. Are you still there or not? We'll talk at the next gathering. [laughs] Now let's talk about Marcel and his training sessions because Bernat and I make a face gels. But my question is, Why not use idem or an agent? Anyone who can do this too? or GPT or Finally, first we have a model hybrid that the whole issue of training, loads, metrics, etc., are their own because LMSs do not They use none, they are not consistent in what long time in that, I mean, it puts you random workouts during and do not maintain Consistency, you have to loads progressive, etc. In other words, we already have a hybrid part that is algorithm and training system engine, which is ours. And then because on top of the LM there is to To give it a behavior is like the LM are pure knowledge, but you have that encode a behavior and this relationship of a human being that you constantly accompanies him throughout the day, that is proactive, that integrates into your weaballs, etc. But in the end, you can do this well. interviews, right? In other words, you can I mean, uh, Chagpt, right? Hey, I want you to be my personal trainer. From Now start asking me what you need to personalize me, my routines and such. Yes, but the first day is fine and It works well because it will make you a training plan, but the second He won't even ask you how, and you won't go. Let's see how you slept, it won't even detect it that you just trained, you won't even see what You've eaten, you won't even remember it What kind of training has he put you on, or what? kilojoules, this is very technical, but How many kilos have you burned during that time? session to tell you, "Eat a little more Today, eat a little less, put in more "Protein, add more carbohydrates." Losms do not have this behavior They'll give you an entrance fee if you do one. An interview, as you say, will make you a 20-week plan, but it already existed. They are the typical PDFs that you can buy. add a Hermes, an idem, an Open Clow, I mean, someone who has your memory, huh? above the CH GPT and the structure, uh think, do the streaming during the night to structure what it is you It's convenient, what have I learned. You can, our experience is one, You can set up a NUA more or less, but The behavior is getting worse and more widespread. Nobody in the sports world knows what Open Club is. EITHER Well, we do know a lot here, but the reality It is the brand distribution. Distribution plays a role important, that's why the part of Cobrandet and the agreement with Valentan, But the reality is also that in our environment, at least in our Open Cloud leaves users very little room for error. far. Which wearables do you work with? With what? Wearables. Talking about Wearables. Uh, we're with Garmin, with Waho which is further. Garmin has a new one, right? Now Yes, it is a bracelet. a bracelet Like the Fitbit, right? And mentioning that We started focusing on cycling. We're going to take out the vertical running in two three months. Then we started to find out because Cyclists usually carry a GPS. Typical brands include Garmin and Yahoo. In Europe has 70% of the market with Garmin Yahoo, but we're also there. in Longtail with smaller brands or even for tracking. And these applications have APIs. Yes, in fact it's quite common. Uh, it The easiest thing usually involves a lot of people Conoceva, which is like a social network where you post, basically it's connected via API to all brands. And how? So, you said, how many Do you have any euros? Did you say 1000 payment? And how much do they pay? €20 per month. The average ticket price is 16 because there is a semester plan and 20 per month. How many have stayed and How many have left? How many have stayed and how many have left gone? Since we started. Okay, if you want, I can tell you the curve of We have more or less retention. The the Month 1 is 88%, We gradually decrease to 50% by month 6 and it's ironed at 35-40% of month 6 to month 17, which is as far as we have so far now. It is considerably taller than the industry. We believe it's because we are the first agent, that is, by the approach of the product, because before there was training apps, uh, that don't have anything that we see and we believe we can improve it lot. In fact, that's what it is, the product. And your main user is a cyclist Or not? Now they are 100% cyclists. And there I was No, but to get an idea, for For example, in my day-to-day life, well I train in the morning, okay? Just get up and I'll tell you, "Marcel, wow, you slept well, I don't know what training had we agreed upon with you because I know you had to wear it this week You have a competition coming up. Take a gel, Yes, or snack on what the gel is. And I told him I say, "Trained this." And he will tell me and then the It detects it automatically, it gives you feedback, you analyze the training together, Discuss whatever you want and from there here, depending on how I've come from Factorial and Jordi tells me, "Let's go eat at an Italian restaurant and I'll make him a photo of the dish." Yeah, Okay. And so, In fact, many people already do. She has the diet planned, so you go with Jordi and you order the dish that he has said NUA, but if you don't, you pass and you can adjust the dinner or adjust the snack. And from here I know because this is true I do it quite often with chatGP. So, what else? What ... happens? What do you do at CH GPT? Do you send it to him? A photo of the dinner or lunch? Yeah. We had a great entrecote, you and I. Today for lunch, uh, precisely and and fries. And what does he say to you? Well, I can roughly tell you then. How many calories can that have? And You do with that, exactly. Of course, you already know what the threshold is. I mean, I do that, I don't do it all the time year, I do it because last year because I have phases throughout the year and I say, well Now let's behave a little better, For example, last year I was in a a month and a half between sick leave, vacation, etc. I said, well, now, so, then Well, I'm going to try to control it a little. that what And do you know how many calories from accumulating more or less so. Yes, a few days is fine. Do it because otherwise, well, I mean It's a brand, isn't it? And then that with GPT chat. So, in the end, Returning to your product, this the The idea has it pretty much figured out because Many people around the world ask GPT chat, hey, I want to get in shape, I want to burn calories, I want to train, I want that. Yes, but the behavior consistent, at least also what we Our users say they end up going NUA because it's much more convenient, more Easy, it's 19 hours a month, which you reduce the GPT premium chat, it's the same and the The product's performance is better. It's more advanced, it's proactive, it integrate your wears that chat GPT does not It happens, it can happen, but Uh, in the end we've coded this coach's behavior that GPT chat does not occur. And I want to ask you a couple of questions like context. My reference is Kalai. Cale, you know it, it's an app, uh, that he managed to bill tens of millions of dollars that the A kid did it, if I'm not mistaken, with 17 years old, all alone, vibe coding, what You took a picture of the food And it told you the calories, right? And from From there, well, with all the billing He hired a couple of his friends from institute and she began to improve and sold. Uh, I think, I'm saying this from memory, I I might be wrong, but I think hundreds of millions of dollars or per or per 100, Let's say, not too long ago. So, You can do what Kalai did. with ShaGPT. Calii's appeal lay in its virality, the distribution and such. So, I I ask, can you explain to us which one has What has been the growth curve of Signaps? No? What does the number look like? records that you have month by month or the approximate metric you have for To understand how that growth is progressing? First, we haven't managed to go viral Yet, we haven't grown much at all bit. We started that a year and a half ago and investing €50, €100 in meta and progressively as the payback is low, because we've captured it very cheaply, from The CAC blended of everything is €15 Pay in a couple of months. €15 of what C, if you do what we've spent in marketing and the users who have have entered and paid and end up paying That's 15 € or by installation, No? For payment, per paying user. Yes, just in case €15 per A paid user account cost €15 but we have spent 18,190 on marketing. I mean, we haven't. scaling. We are now spending 2 €3000 per month in that condition [laughs] that nobody Nobody is scared. It's anthropic, it just release a fitness app after [laughs] having listened. Now we're in, if we do the meta, because For example, to continue with the context of C 40 € we are, well, not this month We spend nothing because we have to pay taxes, corporations, etc. [laughs] Uh, I would also notice that separate it, eh, and lawyers, but we were at €30-35 spending €2,300 a month, but that's only of the pure B2C part. How many signups? Excuse me, how many Signups? In other words, €35 €30,000 is 1000 signups. 20,000 signups in their early twenties signups since signup since Let's begin. The conversion from sign-to-pay It's around 5 and a bit per But how has the issue been doing? How many You have gained new customers in June? In June, well, we went from I think I know it more in MRR, but it was a good month and We were at 13 and we are at 16 in June, That is, €3000 from MR and in May 1000 and something. We're fluctuating too and in April 700, 800 also depends on the And in March 600. It has been progressive each month. But then, to reach 20,000, 1 600 at a time is a lot of months, I don't Which one, because you left in October, right? No, a year and a half. Sorry if I have mistaken. You didn't say October of last year. A year and a half, January 2025. I don't know if I said it wrong, didn't I? Okay. Or maybe the other pitch made me confused. AND From January, from 2500 signups. But I can't get them right, sorry, I can't get them right. the numbers. And how much have you said about MR or AR? 16. 16 of MRR, 200 of ARR. Arox, I think it's a little over 16. Okay. AND And it's all paid for. Recruitment is everything paid. It's the goal. The [clearing] part of B2C is mostly meta and some Google that we also included. And then, Which part is B2C? The current ratio is 80% to 75%. And there's the Cranded which is a revenue share of 70% nua, 30% the sportsman. And is this Decathlon or an influencer? We've only made one influencer. Well, He's not an influencer, he was a champion of world of mountain biking, his name is José Hermida and the idea is to replicate it with more sports figures outside the market Spanup. And this one brought you 25% of the MR? Yes, from users. 250 users paid. Yes, it's a paid service. of 1000. I always do the payment thing, yes. That is, 25%. Yes, yes, yes. Okay, Okay. We move on to feedback. I'll start. If that's okay with you. There are a lot Try it, okay? Well, I'm curious about it, but yes. It's something that I think, I mean, the challenge is the distribution, uh, and the and the and the and the be the goto to. No, I haven't seen much. In the world of endurance racing, there is no... I've seen little that works well in that area. in the fitness, CrossFit, gym section, etc., that he does everything right either. So, hey, that might make sense. Me I'd like to see it, try it, and if in a This week I see that I'm blown away, well Sure, there are 14 free days. [laughs] You have two weeks. Yeah, I know the numbers you're in. I love sharing them. I mean, Now you can see them for yourself from Stripe, so you can see that they are real. Yes, yes. No, definitely, definitely, [laughs] I'm sure I would see it. But, but, but I mean, the problem that What we have is that every time we get involved in B2C we get into [laughs] one We understand the dynamics of B2B, but the The truth is that B2C dynamics are very strange, very volatile, eh, everything changes A lot, isn't it? But on the other hand, well They have a lot of volume, don't they? If you manage Hitting the right key, huh? It can be very interesting. I'm not the target audience, but... I see, I mean, I see value in the proactivity, I mean, what you're saying about him chasing you, No? To say, hey, so-and-so, the thing is that I would turn it off, I would turn it off, that's why I'm not the target audience. There's the option [laughs] too, there's the option of not being proactive. Yes, but then already He doesn't lose it. So I'm not going to pay 20. Of course, it's a contradiction, isn't it? Wow, I find it incredible that 1000 People pay €20. It seems to me incredible. I mean, that's all for that reason alone I want to be more. Yeah, Yeah. I'm less target than Bernat I still don't know what you're doing. with the gel when it's time to train. Hey, Just to confirm, Kalei, that for me It's a reference because it's a case of brutal success, similar, It reached 15 million downloads and between 30 and 40 million in revenue with eh a 20% of, I mean, put 6 Millions in profit, right? And My bought it Fitness Pal 4 or 5 months ago. And that was it. a 17-year-old kid with zero investment because profitability and the hypergrowth, which is what makes it valuable, that's also what made it valuable Profitable, isn't it? So, or whatever he We need to have capital to... grow. So, uh, it's this kind of businesses that, when things are going well, don't They need investment, and when things go badly, they don't. You want to invest in them, don't you? So, and If you are experiencing this curve growth, it has a lot of merit, there is a lot Few people manage to get it and I'm interested to learn more about why you are achieving, but I will have would I'll have this challenge before... to convince me to invest. AND Having said that, excuse me, I remember that we are investors in a space company and First we would have to clarify that no There is no conflict whatsoever with BO, but I agree with what he said Bernard, without being either of them users or potential customers, that Get a B2C Consumer Subscription The app is a miracle, it's one of the most There are difficult ones. And the curve you have Right now is uh enough It's so impressive you want to see more. Perfect. And by the way, the retention curve that You have, if it's as you say, It's bigger than a gym, any neighborhood gym. It's likely. [laughs] We're in cycling, uh, we're not talking about people who go to the gym, etc. I mean, stop. Hey, it's also very niche right now Your target audience, right? They're the biggest geeks. This, as it goes, is going to the more normal population, Yeah. Hey, maybe the shorts will go up or down, I want say, could be. It's not that niche, is it? But yes, more than the gym. Could be. Fernando. Fernando. Uh, I wanted to tell you, there's a person I think I'd be okay with that you spoke, that it is Pablo Villalba. Pablo Villalba is an old friend our serial entrepreneur who founded Efit, which was an app before the world of II, uh, of fitness, of training, of diet and that experienced growth brutal during the early years, He raised a lot of capital and knows a lot from this.
-
2:14:05
, obre el vídeo en una pestanya nova
So, as far as I know, He's in Berlin, but he's from Albacete. I recommend you look for him, and if not I cross paths with you, but it's a person from whom I'm sure you can learn a lot. Cool! We do it. Thank you. Great, Marg. Thank you so much. A lot luck. [applause] We're nearing the end. Before Finally, thanks again to Cactus for help us make these events possible and bet on the community tech And as you who have come know So far, now you have catering like always from the Moons who behave superb with us and so now You can take this opportunity to network and eat a little bit, it's time to have dinner. Bernard, Jordi, we're leaving Vacation, right? Please, to rest a little, a little. We've had a fun year And we'll be back in September. of course. Thank you all. I hope it goes very well. Thank you. [applause] Cactus is an engineering firm located within from the La Salle campus, University Ramón Lulull. Over the years we have formed a knowledge network that puts talent at the service of customer. The [music] engineering of Cactus is made up of five departments: hardware, firmware, embedded, backend and frontend. He The hardware department is responsible for design, produce, validate and certify Custom-made electronics, especially digital systems, analog, communications, control of low-consumption motors and systems. He the Firmware department specializes in in operating systems programming real-time, resource optimization system, energy optimization and [music] Zero Trust systems. He department musted specialist in the operating system optimization Linux for different types of hardware architectures. The department Backen develops architectures measure on different types of cloud infrastructure. He also dedicates to data analytics and to application development. In Front We are dedicated to developing the design and the implementation of different types of applications involving UX, UI and also data analytics, in addition to artificial intelligence. We have several types of clients different markets. Our proposal The concept of value has two main axes: the technical excellence and empathy. All our developments are based on best practices, architectures decoupled, validation and testing, always adapting to the speed, requirements and context of our customers. A competitive advantage of Our company is that we are not a not a typical supplier, but we are one Stop partner due to the wide variety technique that our team has. We accompany our clients from the conceptualization, the design of product, development, validation and certification, production, the deployment and even ongoing support. Cactus Professional engineering [music] services.