Submind YouTube summaries
Thumbnail for Jeremy Howard interview at PytorchCon with Anna Tong

Jeremy Howard interview at PytorchCon with Anna Tong

Watch on YouTube

Video summary

In this fireside chat at PyTorchCon, Jeremy Howard reflects on his early days as a contrarian who championed PyTorch over Google's TensorFlow and predicted the rise of large language models before they were mainstream. He recounts how he and Fast AI initially faced skepticism for investing in an "obscure" open-source tool that felt more human-friendly than enterprise-focused alternatives, only to see their first model built in a simple Jupyter notebook outperform state-of-the-art solutions from major tech giants. Howard emphasizes that the core breakthrough was not just the technology itself but the concept of transfer learning and fine-tuning, an approach that took years to gain traction until it became foundational with ChatGPT. His journey underscores the importance of leading in emerging fields rather than following trends, a lesson he draws from seeing former students now running major research labs and startups because they learned these future-defining skills early on. Howard addresses the current hype around AI agents that promise to automate all tasks for humans, arguing that blindly outsourcing work is dangerous as it leads to skill atrophy and loss of human agency. He warns that while relying entirely on AI might offer short-term speed gains by generating more code, it ultimately results in shipping less actual product because AI-generated code often lacks proper integration and abstraction layers. Instead, he advocates for a mindset where humans use AI thoughtfully to enhance their own craftsmanship, ensuring they remain competent even if the technology evolves rapidly. He shares his personal experience of becoming a better developer by using AI as a collaborative partner that guides rather than solves problems, illustrating how this approach prevents individuals from losing control over their work and avoiding the psychological stress of managing ununderstandable piles of generated code. The conversation also touches on critical ethical issues regarding who controls powerful AI technologies, with Howard insisting that open-source principles are essential to prevent a monopoly by wealthy entities that could destabilize democracy. He draws historical parallels to how societies managed electricity and telecommunications, arguing that just as governments played a role in ensuring universal access rather than letting only elites benefit from these powers, the global community must ensure democratic access to frontier AI models through public investment alongside private innovation. Howard notes ironically that China currently leads in open-source model development due to its strong emphasis on computer science education and openness, suggesting it is not too late for other nations like America to pivot toward similar values supported by pioneering companies like Meta and Nvidia who are already releasing high-quality models under open licenses. Looking ahead, Howard expresses excitement about upcoming initiatives designed to teach developers how to harness AI without falling into the trap of "grind and vibe coding" or relying on autonomous agents that replace human thought. He plans for a new Fast.ai course focused on large language models specifically for coders, aiming to build foundational skills similar to those taught in web programming courses but adapted for the age of generative AI. His ultimate goal is to foster an environment where humans and AI collaborate iteratively within shared workspaces like "Solve It," which applies decades-old pedagogical principles from George Polya's problem-solving methods to modern development practices. By focusing on improving human capabilities rather than replacing them, Howard believes we can navigate the critical transition period of artificial intelligence while maintaining our agency and continuing to create meaningful value for society.
Read the full video transcript
Afternoon features. Here we have a fireside chat with Jeremy Howard uh founding researcher at Fast AI and he'll be in a conversation with uh Forbes reporter Anna Tong. So please warm welcome for both Jeremy and Anna. >> [music] >> here. Should I do here? Perfect. Hi everyone. Hi Jeremy. I'm told by many people here that you need no introduction, but for the two people who don't know who you are here, um Jeremy, you're the CEO of Fast AI. You're the CEO of Answer AI. You created the first large language model and you were the first outside organization to pour resources into PyTorch. Am I missing anything here? >> That sounds good. >> Okay. Well, uh, tell tell us what it was like back then. >> Um, well, it wasn't like this. You know, PyTorch [snorts] was really the baby of uh two people, you know, Sum and and Adam. And it it came from a longer line of software that was written in Lua which is just called Torch. Um and at the time Google you know was investing heavily in something called TensorFlow. >> And TensorFlow never felt very human friendly to me. It seemed very focused on kind of enterprise and >> making things harder for people and easier for computers which felt wrong to me. So when I saw what Sumouth and Adam had done with PyTorch, uh we spent hundreds and hundreds of hours working with it and we were just like this is so much more human friendly and so we just stopped everything else we're doing and we went all in on PyTorch. We were the first to do so. And at the time people were just like you're crazy. You know, everybody knows Google's going to win. Um, and people are like, "Why would I come to your course when you're teaching some obscure open-source random thing?" Because it's better. You should learn this because it's it's going to win because it's better. And you know, the uh the first language model, first large language model, as you say, was a ulm fit that was written in PyTorch. And it wasn't written on some big enterprise thing. It was written in a little Jupyter notebook. and it smashed every state-of-the-art thing that was there. Same problem. Everybody told me, "Why are you working with language models? Like, nobody cares, you know, everybody knows that's not the future." And I said, "Well, they're wrong. You know, transfer learning is the future. This is what's going to work. This is what's going to change everything." And you know it's great because the people who were in those courses where we were teaching PyTorch and we were teaching language models and they did indeed learn the future and now a lot of them are running the major research labs or the CEOs of the big hot startups. You know it's no point following. You got to you got to see where things are going and you got to lead. You know >> what was it like to create the first large language model and how did that how how did that work? So, [snorts] I'd been telling people for years that the natural place for deep learning, you know, the the type of AI that we were teaching and researching was in natural language. And that was extremely controversial. Everybody thought that was stupid. And I got sick of trying to convince people of it. So, I thought I'll just do it. And it was literally by I I decided to do it. And by the next day, I had done it. And the very first thing I tried against was basically the hardest academic benchmark, the IMDb benchmark, the longest documents, you know, the the most tricky um that been very heavily invested in. And my very first result was substantially better than anybody had ever done. And it was just like, holy crap. because I'd studied philosophy back at university and I'd studied this idea that goes back a long time of like what if just being able to, you know, kind of symbolically, statistically, whatever, like finish a sentence, you know, looks like the essence of intelligence. What if you, you know, can you tell the difference between real intelligence and and that? And I thought like, no, you can't tell the difference. So, I've been thinking this for 30 years and now I had like an algorithm that was doing it. But again, it still took a really long time because um Alec Radford at OpenAI saw that work. He told me and he was like, "Oh, this is going to work great." And so he turned that into what became GPT again, you know, PyTorch, Jupyter Notebooks. Um but nobody understood the the core of the essence was this transfer learning fine-tuning idea and it wasn't until um chat GPT came out >> y >> that used that exact approach the exact same three-stage approach that everybody was like oh this is a good idea. So it takes a long time for an idea to you know percolate. >> Yeah for sure. So we've established that you are are contrarian. So let's get into some of your current contrarian takes. So, I think, you know, today we're hearing AI agents, AI agents, AI agents. Everyone is saying the future of AI is for AI to do everything for you. Um, what do you think of this? Do you think that's a good thing or a bad thing? >> Um, I don't think that's terribly likely to be true. But regardless, we can see that there's two possibilities. Either it is or it isn't. Right? The question is, what should you do about it? If it is true, right, and if AI takes over everything and does all the work, then it doesn't matter what you do, you're going to be obsolete. So, whatever, right? On the other hand, um I think it's reasonably very likely that it won't be true and people will be very much needed. Now, what should you do then personally? Well, if people are needed, then if you go head first into agents, agents, agents now, and you like you're going to stop learning. You won't be practicing your craft of coding, of building models, of analyzing data. You'll be outsourcing it all. So, you're going to be in the group of people who just staltify. And I plan to be with a group of people who who uses AI very carefully and thoughtfully to make me continue to improve in my craft. And so I'm using AI now to get better, for me to get better at my work, for me to learn more, for me to get more skills, for me to practice better. Um, so I, you know, for me, I would say for the humans in the audience, you know, you should be focusing on how can AI help you improved as a craft person? How can it help you improve to develop your skills? If you outsource everything, you know, I'm seeing this happening already, Anna. Like people are like forgetting how to do work. You know, they're forgetting they can do work. And if the AI can't do it for them, they're just like they're lost. And the thing is, I think it's bad for the psyche as well to be like outsourcing everything to AI. It's creating thousands of lines of code. You can't understand it all. You can't keep on top of it all. And it's really stressful because it's like then it doesn't work or like later on you have to integrate it and you don't know how and you just get this bigger and bigger pile of debt >> and it just I've seen it weighing on people. I've seen people becoming just depressed that they're no longer competent and they are no longer in control. So to me and the agentic approach is like the computer is in control. The human should have agency. you know, you're not going to go after this and like write up an article where you say, "Please um uh chat GPT, write my article for me." Like you you would be losing your craft as a writer. You know, it's like what's the point? Uh, so I think [snorts] I would be very very careful of spending too much time trying to get agents to do your work for you because I think that is you are on the path regardless of what happens to AI of making yourself uh obsolete and incapable. >> Well, do you think it makes people faster or you know there's this argument that we're now able to write and ship so much more product now? Um, I I'm not sure. I think I think like in the short term maybe, but probably not. The people I know who have been diving deep into AI powered coding, I have personally experienced they seem to be shipping less >> but creating more code. Now, I think the problem is the AI code isn't very good. It isn't very well integrated. It doesn't create layers of abstraction that fit nicely together. When you in when software engineering is going well, you should be getting faster and faster and things should be getting easier and easier. With AI code, it's the opposite. So I people I see people using AI more, create more code, but ship less product. Now sometimes maybe it's a little faster in the next 2 months, but you're not building your skills as well. You're not learning as much. You're not building as powerful a set of abstractions to like craft better and better stuff on. So, in a six-month time frame, I would say almost certainly slower. In a two-year time frame, I think companies that bet too much on AI outsourcing are risking destroying their company because I think in two years time they're going to be looked back and be like, "Wow, in the effort to get a quick twoe result here, we destroyed our competences for an organization to create things that that last So how should we be using AI? Maybe to check our work or to ask questions or how are you using it? >> Yeah. So that's actually so um answer AI which is kind of fast AI's parent organization now is that's what we're dedicated our mission to do is to figure out how can humans use AI to you know where humans have agency and they empowering themselves to sustainably do high quality work and to get the answers they want. So um so we've actually got a a new course coming up where we try to like teach people a way of working with AI where uh you do small pieces highly iterative a little piece and you try to write it yourself and you have an AI kind of in there with you watching and kind of giving you tips and you can ask and but when you ask it's like totally written in a way to say don't solve the problem for them just guide them And we found in this so we've created this uh shared environment called solve it um which is actually based on principles from a 1945 math book by a guy called George Polier um how to solve it. So these ideas been around for decades of like there are principles about how to do high quality work and so we've tried to create an environment and a way of thinking where the AI is in there with you >> interacting with you and the environment that you're in. Um so my the similar ideas also my co-founder Eric Ree created the lane startup a book and movement >> and the whole approach of you know the minimum viable product of short iteration cycles it's it's all the same thing. So he's writing his new book with the help of Solvent. Solvett's written none of the words, right? But it's in there with him in the book as he's writing it, helping him fact check, helping check transitions, you know, and and and the more he writes with it, the more it knows what he's trying to do, what kind of help he needs. So again, he's he's becoming better and better as a writer because he's practicing his craft. He's getting the feedback and he's doing the work, you know, and he's the agent. >> Yep. >> Um, and same for me. I feel like as a developer, I'm a much better developer than I was 2 years ago cuz I'm all about using AI to help me get better. I want to I want to outrun the AI. Like if in 3 years time AI is a lot better than it is now, but it's still not making all of us obsolete, then which people are going to be valued? It's going to be the super great people, right? So if you're somebody again who's let AI do all the work, you've basically become incompetent. You're no longer your own agent. You're not going to be one of the people that still matters. So, you know, as AI gets better, it's more and more important that you are that your skills are growing faster than the AI skills. >> I gotcha. So, um you've been a major open- source advocate in the past. Do you still feel this way with um frontier level AI models? Should they be open sourced? Yeah, I mean I I never used to be a particular kind of radicalized open-source advocate or whatever. I mean, I used it a lot. I loved it. Fast mail, which is one of my earlier companies, big email provider that lots of you probably use. We were one of the very first to use Linux. We were using Linux at a larger scale than anybody in the world at that time back in the late '90s. Um, I've always felt like open source is, you know, the right way for the global community to advance our software capabilities. But now I think it's much more important. Um, AI is now a source of power in the world. It's becoming an increasingly important source of power. There have been many times throughout human history where technology has created power. various types of technology, you know, everything from the printing press through [snorts] to, you know, technologies like effective education or writing. At every point there's been a major new source of power, a group of people says like that's too dangerous for most people to have because some people will misuse it. So only the rich and powerful should have it. And we're seeing it again now with AI. Mhm. >> The problem is that actually the rich and powerful are not the people that could be most trusted with that power. So if we let AI just be in the hands of rich and powerful people, the best AI, it it could literally tear apart democracy at its foundations. So I feel like we're actually at a point now where we have to reinvest in enlightenment principles. Enlightenment principles is basically this idea of saying okay there are bad people in the world but we believe that that humans overall are are good. They are a force for good and that when there's a new power in the world distributing it is the safe thing to do. Centralizing it is the unsafe thing to do. So open source is the way we ensure it remains distributed. You know, PieTorch is what everything is built on today. You know, none of this would exist if this open-source didn't exist, you know, and so I think we all have to recommmit to this principle that okay, some people will use this technology to do bad, but most people will use it to do good. As a result, by using this technology to help defend against the bad guys will be much more powerful than locking it away so that only the rich and powerful can use it. >> I mean, I think AI is different though because it requires vast amounts of compute in order to train frontier level models these days. So, what should we do to, you know, ensure that we still have democratic access to AI and that, you know, small labs can still make great models? >> Yeah. Again, you know, we've faced all this before as a global community and countries that have looked at things that inquire huge capital investments like giant coal fired power plants, you know, um, uh, fiber optic lines down every street for telecommunications and so forth. We figured out ways to do it. We've said like, okay, the government needs to play a role in ensuring that everybody has some level access. um private institutes, you know, have a role in taking advantage of markets and capitalism, you know, in that as well. And like we need to find a way again of um not ending up with, you know, 10 different groups all spending a trillion dollars. You know, the government needs to do some pieces of it. The government needs to ensure that there's some level of access, but there also seems to be some level of competition. Um I I don't think any of this is new, right? Like we you know the the early days of electricity, some companies, countries like China only gave electricity to the elite and some countries like America said we're going to put it down every road in New York and make sure every everybody has access to it. And again, I think in the end we found that trying to make sure that everybody has access to these technologies is the way that actually benefits us all better. >> Do you think it's too late? I mean we are already in a state where just a few companies have a lot of have all the access to compute >> you know I mean it's funny to say this given the kind of geopolitical history but at the moment China's the country that's saving us from that so the best open- source models today are all Chinese um I don't find this particularly surprising because you know having spent quite a bit of time in China is a very you know the kind of the system there really invests in computer science science and in math and a lot of people believe in in openness. Um so that's been very important but I also think yeah it be great to see America turn things around the one company actually that has stood out head and shoulders above the others. than that there's two uh one is Meta right >> the creators of PyTorch and the other is Nvidia who just in recent months have created some of the world's best models and they are open source and they are openly licensed so no I don't think it's too late I think there are some pioneering companies that are showing the way >> so final question for you um what's something you're really excited about in the next year that's that you think will be enabled by AI or a couple things you're excited I mean, I I'm honestly really excited about the work we're doing. You know, I feel like we've discovered a way to work with AI that is deeply human, uh, deeply supportive and is the very opposite of grind and vibe coding and agents, you know. Um and it's small iterative steps where the you know humans and the AI are kind of in the same canvas working together and the focus is on improving the the human capabilities. So I'm sure a lot of people in the audience have done the fast AI courses in the past. We're going to have a new fast AO course next year uh which will be language large language models and deep learning for coders and it'll be really like harnessing this idea of like getting everybody in this environment to to learn the foundations. You know, we got to do a similar thing for kind of the foundations of the internet and web programming. We'll do a similar thing for um building startups. Uh, I'm I'm really excited about this. You know, I feel like we've discovered something important here at a really critical time of how to work with AI in a way that that supports humans rather than replaces them. >> Well, thank you very much for your time, Jeremy. >> Thank you, Anna. [applause]