The Exchanges

Q We're gonna be talking, I think, about both, um, your experiences in research and academia, and then we'll also separately be talking about Together, which is a company you're involved with now. Um, could you tell us a little bit more about what the center does and what you're focused on?

A Yeah, so the Center for Research on Foundation Models, uh, started two years ago, is under the Human Centered AI Institute at Stanford, and the main mission of the center is, I would say, to increase transparency and accessibility to foundation models. So foundation models are becoming more and more ubiquitous, but at the same time, one thing We have noticed is the lack of transparency and accessibility of these models. So if you think about the last decade of deep learning, it has profited a lot from having a culture of openness with tools like PyTorch or TensorFlow, data sets that are open, people publishing openly about, uh, paper, about the research, and this has led to a lot of community and, uh, and progress, uh, not just in academia, but also In industry with different startups and hobbyists and whoever just getting involved. And what we're seeing now is sort of a retreat of that open culture where models are now, uh, being only accessible via APIs. We don't really know all the secret sauce that's going behind them, and there's sort of limited access.

AI assessment note: “the main mission of the center is, I would say, to increase transparency and accessibility”

Q So, but you've also taken a point of view that we should, you know, expect superhuman. If we, if we see superhuman performance from these models, like Holding them to the standard of a human doctor is actually insufficient as well, right?

A Yeah, I, I think that's a, that's a great point, is that for ages, human level has been the target for, for AI. And that has really been kind of a north star that has fueled many dreams and efforts and so on over the decades. But I think we're getting to a point where along many axes, it's a superhuman or should be superhuman. And, and I think we should maybe define more of an objective measure of like what we actually want. We want something that's very reliable, is grounded. Uh, you know, I often want more statistical evidence when I speak to doctors and sometimes fail to get that and have something that would be sort of a lot more principled and rational. And, and so this is more of a general statement about how we should think about technology, not just Chasing after mimicking a human because we don't have a lot of humans and yeah,

AI assessment note: “we're getting to a point where along many axes, it's a superhuman or should be superhuman”

Q I guess I'll ask two questions. Um, one is if people wanted to participate in together, is there a client they can download and install or use, or how, how can people help support the together efforts?

A Yeah. So we are developing a client, um, that, uh, will be made available both from the perspective, joining the together clouds so that you can contribute your compute, but also where we have an API that we're developing so that people can use, um, Um, the, the together infrastructure to do inference and fine tuning, um, models. Um, we are also training some open models. So we have this, um, something called open chat, uh, kit that's, um, uh, we're releasing soon. And this is built on top of Eleuthera AI's, uh, Neo X model, but, um, You know, improve to include various different types of capabilities. Um, it's still a, you should think about it as really a work in progress. What we've trying to do is open it up so that people can, um, play with it, give feedback, and have the community improve this, um, together, um, rather than us trying to produce some finished product and putting it out there. Um, this goes back to the point about involving You know, it's the spirit of open source and involving the community to build, um, these, uh, foundation models together as opposed to someone unilaterally, uh, building them.

AI assessment note: “So we are developing a client, um, that, uh, will be made available”

Q What do you, what do you think are some of the more interesting research programs right now in that area?

A I think that It's, it's interesting to, uh, remark on what's happening because to a first order approximation, larger models trained on the relevant data seem to do well on various benchmarks. I think that maybe there isn't enough emphasis on Data efficiency and how quickly you can get and how robustly you can get to these points, because we know, uh, it has been well documented that benchmarks can be gameable. So even though you do on a benchmark doesn't mean you've necessarily solved, um, the problem. So I think one has to be a little bit cautious about that. So obviously scale and more data is just one clear direction, but in terms of orthogonal directions, what are the methods? Several things have to happen. One is, uh, we have to have ability to handle greater context lines. If you think about a long reasoning chain, you know, transformers are fixed, um, and there's ways to extend it, but fundamentally it's sort of a fixed model. Um, there's, let's say advanced problem solving. For example, if you want to solve, um, uh, a math problem, you'll improve something. The language model generates sort of thinks out, uh, this chain of thought and It generates token by token, and then it generates something. But we know that humans, when they solve a problem, you try different things, you backtrack, there's, it's much more flexible, iterative, um, and it can last a lot longer than …

AI assessment note: “in terms of orthogonal directions, what are the methods? Several things have to happen.”

Q Actually, it will ask if you'll risk making a prediction that is impossible. Um, emergent behaviors of models at the next level of scale, anything you might predict? Emergent capabilities, if we wouldn't have thought chain of, uh, chain of thought or in context learning would work.

A I, I can give you an example of something I think is emerging, and I can give you an example of a hope, but I don't know if I would call a prediction. So what we're seeing today is The ability to instruct a model, um, using natural language to do certain things. You see a lot of this online with ChatGPT and BingChat where you can just, and some of Anthropic's work as well, you can instruct a model, uh, to be succinct, um, generate three paragraphs in the style of, and so on. You can lay out these guidelines and have the model actually Follow. So this instruction following ability is getting extremely good. Now, I will say that How much is emergent and how much is, ah, not? It's hard to tell because, um, a lot of these models, it's not just the language model that's trained to predict the next word. Um, there's a lot of, you know, secret sauce that goes under the hood. So, and if you define emergence of, you know, it was not intended by the designers, I don't know how much of that is emergent, but at least it's a capability that I think is very striking. Language models currently mix stuff up. Um, they hallucinate, um, and this is clearly a pig problem. Um, and almost in some ways a very difficult problem to crack. The hope is that as models get better, that some of this will actually go away. I don't know if that will happen, um, but, but I think that would be extremely nice be…

AI assessment note: “I don't know if I would call a prediction.”

Q How far away do you think we are from that? Is that something we can do today? Is that two years away? Is that five years away?

A You know, these, um, projection questions are extremely hard these days. Um, I think you can, Do, uh, limited things, um, already in terms of doing literature search, it can generate things. Um, and I think you're at the level where it could probably generate things and, you know, I, I think it would still be a lot of, you know, human loop, but you could generate probably, uh, let's say, uh, I don't know, a class project, uh, type of project. Um, could it really do something completely like a breakthrough? That seems maybe harder, but on the other hand, AlphaGo was able to discover completely kind of alien different strategies and with the right, you have to set it up correctly. I don't think you can just generate from a language model, but if you set it up properly, maybe these models can actually discover new things. I remember reading a paper, even maybe there's like five years ago where a bunch of material scientists used, um, just word effects. Which is just work vectors from over 10 years ago, and they were able to discover new, um, you know, thermodynamic properties of, you know, materials. And I imagine that today with a much more powerful models, um, you should be able to do, you know, a lot more than that.

AI assessment note: “projection questions are extremely hard these days. Um, I think you can, Do, uh, limited things”