YouSaid · the spoken record

Alex Lebrun

lines on the record
78
first
2023-06-19
most recent
2023-06-19
sittings or episodes
1
sources
podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

  1. “So obviously, I think the fundational model that we'll win will be open. But that being said, we should be careful because even an open model trained with open data to me is not that open because when you have like 300 billion parameters and it's a huge black box, you don't understand why the output is what it is. Is it really open? And so I'm just putting a little bit of salt on what I hear and like open models or you understand everything. It's explainable. It's predictable. It's not true. An open model can be as hard to predict and that's a closed model.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  2. “The reason is simpler than that. The reason is nobody could predict that LLMs would be so useful and powerful before you train one at this scale. And who in the Google org chart had the incentive to invest $500 million and just to see this without any business benefit for the company? If you add to that the legal department who is not that happy that you are releasing random chatbots, we can say anything, then nothing happens. And this is what happened at Google and probably also at Meta where nobody has the incentive to do that. And so OpenAI as a private company, they need to show something to build products and they have this curiosity and the financial power and they did it.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  3. “Fair, you know, some of them can move fast, and it's not fair to say they are slow, and that's it. I think the income and suffer from problems because in many cases they will do AI enhanced features. And this is what all the ones you mentioned did, the notion you still have a document, you edit, you have a cursor, you write hello, by the way, you can call ChatGPT and to summarize the paragraph. It's what I call spreading a little bit of AI dust on the magic dust on your existing product. Woodos are differently you can think about building knowledge for your company. Probably something will come and destroy Notion and Google Docs and all of them with a totally new paradigm that is made possible by AI. Certainly these incumbents won't do it. So I think the best incumbents can benefit from AI to be competitive in their existing markets, but I don't think they will invent the totally disruptive thing that will kill them, you know, Kodak. Kodak invented them.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  4. “So incumbents have a huge advantage through distribution. That's the one thing that is hard for us as startups. But incumbents have many disadvantages. I think first they are very slow. You mentioned Adobe. I was sure you would because this is one example that everybody has, but then who else was as fast as Adobe? So I think most of the incumbents won't move fast. And in many cases,”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  5. “Make a successful financial model or company, you need incredible scientists and a huge amount of money. And if you check these two boxes, I think you can still do it. And you need focus. Google and Mela could have done what OpenAI did easily and they didn't because it was not the priority and there is only legal people around. But I think a new startup with enough talent and enough money. And when I say enough money, I'm talking about billions. Just to be clear, then I'm sure you can do at least as well as Pan AI.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  6. “We probably have most of the existing companies, maybe a few will be started. There is a lot of hype and it's public, so I can spill the bean about Mistral, a new one created by three engineers, two of them from Facebook. And I know them. And you can still build this financial model.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  7. “And everybody still is covering new things. But three weeks ago, there was a paper about Lima. Is this Lima paper shows that with only 1,000 questions and answer examples, so very, very small data sets, they get something used for fine-tuning, so the second stage, they get something that performs better than GPT-3 and almost at the level of GPT-4 with only 1000 QA for fine-tuning. So the thing with LLM is it looks like a huge pre-training with text. It's learning mostly through pre-training with huge amount of text and then with a small amount of data with high quality data, you can get a lot, you can get what you need and train your model to do what you expect with very small data sets.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  8. “So, a large language model is trained in two phases. First phase is unsupervised pre-training. So it's just fed with huge amount of text and it's trying to predict the next world. And so this is what we call unsupervised pre-training. Unsupervised because you don't need any human intervention, the text by itself is enough because it will try to predict the next word and on and on and on. And then the second phase is fine-tuning when you use different machine learning techniques. So for instance, reinforcement learning for chat GPT, where you train your model to follow instructions like front GPT is when I ask a question, you should tell me something that looks like a right answer or looks like a reasonable answer to what looks like this question. And so the fine-tuning is the second phase where you give more precise instruction on how it should behave. To do that, yeah, there are several techniques. So everybody is still learning, you know, it's like huge bag boxes.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  9. “I think VCs are always one train late. So having a lot of proprietary data was very, very important for the last cycle five years ago. Maybe it's less and less true. You need some to bootstrap your model. So for instance, at Nabla, we build a dataset of 30,000 medical consultations with patient consent. We have to pay doctors. It's very hard to build this because we need this to bootstrap our product. But then with the new pre-trained”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  10. “Absolutely agree with that. Will you drive the car, you drive today in 10 years? I don't think so. There is so much progress so fast that I don't see why we would use the same in one year.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  11. “Mean by construction, the LLM has to output something. And so if there is nothing to say, make something up that looks natural. So this is probably what Emad meant by design. He cannot not output something. And so this will be an hallucination.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  12. “How it will work, of course. You have every two weeks incredible things coming out of research. And if you just assume it's a static in a year from now, your product will be very dumb compared to your competitors' products or more expensive to run or very slow.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  13. “I think I have because I know how it works internally. So I know the limitations when something is wrong. I know where I should poke to find a solution. I also know when there are many, many and more and more elements available which one will probably be the better suited to my problem. And I know what kind of machine learning I have to build around by LLM to balance for the weakness of it.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  14. “Existing users and products. There is a new infrastructure, there is a new game, new rules. The companies who are the faster to understand these rules and play along these rules, we win and build the best products.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  15. “I don't think it's fair when the C language came out around 1972. Some people said, hey, now it's so easy to build software. Every software is just a thin layer on top of C. And of course, it's not true. And we know it in the same way when databases came in early in the 80s. So it's not fair to think that any AI application on top of LLM is just a thin layer. LLM is a new kind of resource. It's like a new kind of infrastructure that everybody can access, sure. But then it's completely novel. It's very hard to control. You have hallucinations. It's non-deterministic. It's very hard to configure. You have several knobs you can change between fine-tuning or how you prompt it and many other, it's very hard to control. And also it changes every week. You have new LLMs coming out that claims to be better than the ones from the week before. How do you decide when to change your LM? How do you do it without breaking your”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  16. “From our standpoint is really, really good. We've been building AI companies for 22 years, so I've seen this cycle several times. So after a while, you're not surprised anymore. The only thing you need to know is when to add or remove AI from your deck and it will change about every three or four years.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  17. “For the general public, it looks like a very big step function with huge advancement every 10 years. I think from the inside, it's much more continuous. So, for instance, ChatGPT is based on GPT-4, which is based on GPT-3, was released three years ago. GPT-3 is a large language model. It was invented before that. Transformers where the paper was released in 2016. The progress is continuous, but the public perception of it is very, very discontinuous.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  18. “The solution first, I think, is with external people. Even my investors were too close to me, but if you go to people farther away, they have less, who are very successful and they have no incentive to please you and you get more information, more data from them. And internally, it's a matter of just telling your team, okay, you should challenge our decisions. And we did that a lot at Nabla. And sometimes too much, you know, they challenged everything after that. Also, after you make one or two mistakes, the team will more naturally come forward. And this is before you reach the right level of challenger.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  19. “So, when we decided to start in healthcare, we were an engineering team very strong in AI and engineering, but not so much in healthcare. And we made actually a good decision to start as a B2C healthcare company to learn. But then we lost our way a little bit and we went too far before realizing that it was not the right path and that we should go back to a B2B business model. We probably took too much time to make these decisions. It's just an example.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  20. “Sub getting easier, you learn some lessons, you don't do the same mistakes again, hopefully, but get new dangers, you know, new issues. The one I suffered from when I started Nabla is what I call the Kim Jong-un entourage trap. You know, you see this video of Kim Jong-un and he's visiting something and all the generals are around him with a notebook and everything you say will be written down if you fail to do that. I guess you disappear the next day. And nobody, of course, will ever, ever say, I disagree with you or challenge any of these decisions. And starting my third company after two exits, I felt like my investors were always agreed, my team always agreed, people around us. And so I was not challenged enough. And we made some early mistakes that probably were not challenged enough.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  21. “Say why my ID would work. And I arrive at the meeting and first minute he tells me, Alex, I agree with you, it will work. So instead of hiring 200 concierge, let's hire 100,000 and we will kick Will tomorrow. And then I completely reverse my course and say, no, no, no, Mark, I actually am not sure it will work. It will be crazy to hire 100,000.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  22. “Many things. One thing I like, he spent not a lot of time outside the company. He had like 30 minutes meetings back to back from morning to evening. And he prepared really well every meeting. So you have to send in advance like a short note about what decision you are expecting from him. If you fail to send this document 24 hours by the mute before your meeting gets cancelled. And so in many cases you arrive in front of him, he's read your document, he's gathered all the information, the data he needs. So the meeting is very fast. He will challenge what you ask for maybe and make a decision. So this way to work, I think there is much to learn. I give you an example of that. When I wanted to hire my 200 human concierge, my reasoning was we will train this AI. It will work really well and we will steal business from Google in terms of search law, like exactly what ChatGPT may do in the time to come. I prepared answers to every possible objection. He could object about everything I was ready to.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  23. “So, first thing that amazed me when I arrived on Facebook is I thought all big companies were slow. And actually, when we arrived at Facebook in early 2015, there was maybe 6,000 employees and it was going so fast. You know, it was a machine, very well oiled machine. engineering was so efficient. I learned it's possible to be that big and very, very efficient. And when you are in this region, there is nothing you cannot do. Just after I joined, I started to work on a project and I wanted to hire 200 people to a human concierge to train my bots in real time. It was not in the budget, of course, but a week after I had a meeting with Mark and in about 10 minutes, Zuck challenged a little bit my idea and he said, okay, let's do that. And I could hire 200 people in Menlo Park. And when you have this PED and agility and unlimited resources of a company like Facebook at the time, I was really shocked by this at Facebook.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  24. “Where they go on the roof ali, and there are these people doing barbecue and say, yeah, we are vest and rest. And I saw myself in this vest and rest situation. And after four years at Facebook, I suddenly realized, okay, it's time to go out and back to the arena. And we've learned so many things that Facebook AI research is time to try to push these things to the real world.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  25. “Was sitting at Tara at Facebook headquarters in Menlo Park. We joined Facebook through the acquisition of my second startup. And sitting at the Sarah with cocktails looking at the sunset, suddenly it reminded me of a scene in Silicon Valley, you know the show.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  26. “When we started, I thought childbots would become very, very intelligent after three, four years and get to AGI and replace humans in call centers. And the more I worked on the problem, the more I realized it was really, really difficult to do that. Actually, the first time we released about in Europe was in 2004 to the French railway company. And we were very, very happy to have this customer and the deputy CEO tried it. And she asked, my name is Wong on the reservation. Our bot answered hello, wrong on the reservation. And then we realized, okay, there is still haraf work. It's not smart at all. And so the more I worked on these things, the more I realized it's not ready yet. So suddenly for the last year, two years, we reached a point where market timing might be finally finally good, you know, where the product is ready and people have evolved to maybe 10 years after Siri, people have changed. So maybe the market timing is now.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  27. “So 22 years ago, I fell in love with a chatbot. Her name was Sibel. And really, really, I was on summer night in Paris in my basement, alone with my computer. And I Up to you. I was mesmerized by this chatbot, you know, the fact that the machine is trying to understand you, language, and can generate some words. And I was really, really struck by this thing and I decided, okay, I will spend my life building chatbots. And so I founded a company doing customer service bots 22 years ago, very early. And this is how I started this series of companies in the domain.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source

  28. “So, the new regulation is a disaster. It means 100% of the LMs that were trained these last three years would be illegal in Europe. It's not connected to the reality. The intent is good, but the limited put on how to train a model and how to operate a model makes it in practice compared to what we do today, makes everything illegal.”

    2023-06-19 · The Twenty Minute VC · 20VC: Why No Models Today Will Be Used in a Year, Why Open Will Always Beat Closed in AI, Why Proprietary Data is Less Important Than Ever And Why EU AI Regulation is a Disaster with Alex Lebrun, Founder & CEO @ Nabla · IDENTIFIED FROM THE TRANSCRIPT · source