No Priors Ep. 46 | Best of 2023 with Sarah Guo and Elad Gil

1. 0:00 – 0:27

   Introduction

   1. SGSarah Guo0:00

      (instrumental music) Hi, No Priors listeners. Happy 2024. This week, we're taking a look back on 2023 by bringing you clips from a few of our favorite conversations of the year. We had so many insightful guests and these are really just scratching the surface. We'll list all of the episodes featured, so you can go back and re-listen to the whole conversation. Up first, we have a clip from our conversation with Ilya Sutskever, the co-founder of OpenAI.

2. 0:27 – 3:11

   Ilya Sutskever on the cap profit model

   1. SGSarah Guo0:28

      We talked with him before all of the drama with the board asking Sam Altman to step down, and then his return, so we don't touch on any of that. But in this clip, we talk about OpenAI's nonprofit roots and their evolution into the capped profit.

   2. ISIlya Sutskever0:42

      So the goal of OpenAI from the very beginning has been to make sure that artificial general intelligence, by which we mean autonomous systems, AI that can actually do most of the jobs and w- activities and tasks that people do, benefits all of humanity. That was the goal from the beginning. The initial thinking has been that maybe the best way to do it is by just open sourcing a lot of technology. We later, and we also attempted to do it as a nonprofit. Seemed very sensible. This is the goal, nonprofit is the way to do it. What changed? At some point at OpenAI, we realized, and we were perhaps among, among the f- the earlier, the earliest to realize that to make progress in AI for real, you need a lot of compute. Now, what does a lot mean? The appetite for compute is truly endless, as, as now, as it is now clearly seen. But we realized that we will need a lot, and a nonprofit was, wouldn't, wouldn't be the way to, to, to get there, wouldn't be able to build a large cluster with a nonprofit. That's why we became, we converted into this unusual structure called capped profit, and to my knowledge, we are the only capped profit company in the world. But the idea is that investors put in some money, but even if the company does incredibly well, they don't get more than some multiplier on top of their origin- original investment. And the reason to do this, the reason why that makes sense, you know, there are arguments that... One could make arguments against it as well. But the argument for it is that if you believe that the technology that we are building, AGI, could potentially be so capable as to do every single task that people do, does it mean that it might unemploy everyone? Well, I don't know, but it's not impossible. And if that's the case, it makes sense, it will make a lot of sense if the company that built such a technology would not be able to make, uh, infinite... Would not be incentivized, rather, to make infinite profits. I don't know if it will literally play out this way because of competition in AI. So there will be ma- mul- multiple companies, and I think that will have some unforeseen implications on the argument which I'm making, but that was the thinking.

   3. SGSarah Guo3:07

      Up next, we have a clip from our conversation with Alyssa Henry, the former Square CEO.

3. 3:11 – 5:25

   Alyssa Henry on how AI can small business owners

   1. SGSarah Guo3:11

      We talked about how AI can help small business owners with all the complexity of the parts of the business they don't love.

   2. AHAlyssa Henry3:18

      What's so exciting to me about, um, kind of really how the landscape has changed and the, the, the technology advances in the last year are how much better the tools have gotten a- and how much more broadly applicable they are in terms of bringing, kind of, expert assistance to much larger audience, right? But it, it effectively unlocked the consumer and started to then show what this technology could do, um, when then, you know, further integrated into domain-specific areas. You know, you go talk to small business owners, most of them will tell you, "Gosh, I know I should be doing marketing," right? Like, "I, I know I, if I was more effective in doing that and reaching out my customers, you know, I could drive more business, but I gotta tell you, you know, I work all day and then I come home at night and I gotta take care, you know, take care of my family. And then it's 8:00 PM and I'm starting to think about, gosh, you know, do I just need to chill for a minute or, you know, am I gonna spend the next three hours trying to, you know, create an image and write text for the campaign and everything like that?" And what they'll tell you is like, "I know I should be doing this stuff, it's just too hard and it takes too much time and I'm not an expert. Like, I got into doing this 'cause I love cupcakes, not because I like writing email marketing," right? Um, and so what's exciting about all this technology, you know, that's one example, but there's so many of these kind of different things where, um, just the, the ease of use and the accessibility opens up what previously was effectively just massive white space, right? It was customers or people that if it was easy enough to use, if it was accessible enough, if it was cheap enough, they go, "Yeah, that would be, that would be huge for me." But it was, wasn't accessible, it was too expensive, it was too hard to go find and hire a marketing consultant to do it for me, and the ROI wasn't there and blah, blah, blah. So I think this, this, you know, the evolution that's occurring right now is, is exciting in part just because of really the h- you know, previously unaddressed demand that it's unlocking.

   3. SGSarah Guo5:21

      We also talked to Mustafa Suleyman, the co-founder of DeepMind,

4. 5:25 – 8:53

   Mustafa Suleyman on defining intelligence

   1. SGSarah Guo5:25

      and now co-founder and CEO of Inflection AI, about how his team works to define intelligence and emotional intelligence and give themselves measurable benchmarks to move toward when building their models.

   2. MSMustafa Suleyman5:36

      Spent a lot of time with Shane Legg as well and Shane was really the core driver of the ideas and the language around artificial general intelligence. I mean, he had worked on that for his PhD, um, uh, with Marcus Hutter, um, on definitions of intelligence. I, I found that super inspiring. I think that was actually the turning point for me, that it was pretty clear that we at least had a thesis around how we could distill...... the sort of essence of human intelligence into an algorithmic construct. And it was, it was his work in, I think he, I think for his PhD thesis, he put together like 80 definitions of intelligence and aggregated those into a single formulation, which was how do we, um, you know, the abil- the intelligence is the ability to perform well across a wide range of problems. And he basically, you know, gave, gave us a measurement, an engineering kind of measurement that allowed us to constantly measure progress towards, you know, w- whether we were actually producing an algorithm which was inherently general, i.e. it could do many things well at the same time.

   3. SGSarah Guo6:45

      Is that the working definition you use for intelligence today?

   4. MSMustafa Suleyman6:49

      Um, actually, no (laughs) . I've changed (laughs) . Um, I, I think that there's a more nuanced version of that. I th- I think that's a good definition of intelligence, but I think in a weird way, it's over-rotated the entire field on one aspect of gen- of intelligence, which is generality, you know? And I think, um, OpenAI and, um, then subsequently Anthropic and others have taken up this default sort of mantra that, like, it, all that matters is can a single agent do everything, you know? Can it be multimodal, can it do translation and speech generation, recognition, et cetera, et cetera? I think there's another definition which is valuable, which is the ability to direct attention or processing power to the salient features of a, of, uh, uh, an environment given some context, right? So, um, actually what you want is to be able to take your raw processing horsepower and direct it in the right way at the right time. Because it may be that a certain tone or style is more appropriate given a context. It may be that a certain expert model is more suitable, or it may be that you actually need to go and use a tool, right? And obviously we're starting to see this emerge. Um, and in fact, I think the key, and we can get into this obviously in a moment, but I, I think the key element that is going to really unlock this field is actually going to be the router in the middle of a series of different systems which are specialized, some of which don't even look like AI at all. They might just be traditional pieces of software, databases, tools, and other sorts of things. But it's the router, uh, or, or the kind of central brain, um, which is going to need to be the key decision maker. And that doesn't necessarily need to be the largest language model that we have.

   5. SGSarah Guo8:47

      Up next is a snippet from a recent conversation we had with Reid Hoffman. He's talking here about how we should think about the risk

5. 8:53 – 11:47

   Reid Hoffman’s advice for co-working with AI

   1. SGSarah Guo8:53

      of labor replacement, and how people can make a plan to best work with AI.

   2. RHReid Hoffman8:58

      I mean, the obvious thing about AI that everyone probably listening to this podcast already agrees with is that it's somewhere between the largest, you know, tech transformation of our lifetime and perhaps the largest tech transformation of, of, of human history. And one of the things I use to describe it is like steam engine of the mind. So just like the steam engine gave us physical powers, you know, kind of superpowers of, you know, construction and transport and manufacturing and a bunch of other things, this will give us a whole bunch of mental superpowers. It's both the implication of humanity, um, which is part of what the Impromptu book was gesturing towards. And also there will be some places where we will create, you know, kind of, um, uh, substitution, uh, replacement of work in various ways. And obviously we'll get into some depth on that. But I think that's the, the, the macro picture. And then with that, of course, there's tons of things that are current status and current needs. And, you know, I think everyone tends to a little bit over-predict, like, how quickly things, like, everything will change next year, and that's not gonna happen. Um, but then they tend to under-predict, you know, 10, 20 years, um, in some ways in terms of how the transitions... Although, you know, obviously because just like all technologies, the doomsayers come out first. Um, whether it's the printing press, electricity, everything else is like, "This is the end of the world." You can go back and you can find that this is the end of the world in each of these things. You know, the printing press was described as, as degrading human capabilities through cognition and spreading misinformation, um, as, as an example. And, um, but you know what I'd say that probably as an arc, the thing that I would want to see more of in the, and that's part of the reason why I did Impromptu the way I did, in the creation, theorization, and the design of what we're doing in artificial intelligence, is more in the kind of, um, symbiotic, uh, amplification loop. We tend to, as technologists, say, "Well, I'm gonna have autonomous vehicles and they're gonna drive separately," which I think is a good thing in that case, uh, because I think, you know, you don't need an amplification loop. You just need, uh, effective logistics, you know, safety, uh, you know, save the 40,000 deaths that we currently have in, in human-driv- driven vehicles and so forth. You can go in depth in that if that's useful. But like, like the fact is there's gonna be a whole bunch of things that are actually gonna be better with people plus, um, AI. That plus is a thing to focus on, and I think we haven't nearly as much. And that's, of course, part of the reason I wrote Impromptu.

   3. SGSarah Guo11:39

      Our conversations on No Priors can range from the philosophical to the extremely practical. Our conversation with Daphne Koller from insitro was a look into how

6. 11:47 – 13:15

   Daphne Koller on probabilistic graphical models

   1. SGSarah Guo11:47

      AI can improve the economics of biotech discovery. In this clip, she's talking about probabilistic graphical models as a precursor to current architectures.

   2. DKDaphne Koller11:56

      So I think that, um, just like, like in most fields, there is a swing of the pendulum. A lot of, uh, the early work in probabilistic graphical models was hugely influential in bringing, um, artificial intelligence more into the world of machine learning and, uh, and working with numerical data rather than just symbolic AI. Um, and then I think the advent of, um, deep learning, uh, pushed that to the side a little bit because there was so much power that could be gained from basically the kind of pattern recognition, um, from raw inputs, um, raw images, text and so on, without having to worry very much about interpretable representations. What I think we're starting to see right now is a, uh, the pendulum starting to swing back in the sense that there's a greater understanding that you really need a bit of both. You need that, uh, hugely powerful pattern recognition that we get from deep learning, but you also need the ability to reason about things like causality and you also need some interpretability of your deep learning model so that you can potentially convey to a clinician why you made the decision that you did. And so what we're ending up with, as, uh, a really powerful paradigm is some kind of synthesis of the ideas for both of these disciplines coming together.

   3. SGSarah Guo13:11

      Next, we have a clip from our episode with Noam Shazeer, the celebrated

7. 13:15 – 14:27

   Noam Shazeer on the possibilities of LLMs

   1. SGSarah Guo13:15

      Google engineer, and now the co-founder and CEO of Character.AI, where he talks about why he's a text nerd and the possibilities of language models.

   2. NSNoam Shazeer13:24

      I've just had my head down in, uh, (laughs) in language-

   3. EGElad Gil13:27

      Yeah.

   4. NSNoam Shazeer13:27

      ... like the, like here you have, like, something that, like a, a problem that, like, can do, like, anything. Like, I want this thing to be good enough, so I just ask it, like, "How do you cure cancer?" And it, like, invents a solution. Um, a- and, you know, like, so, so I've been totally ignoring, like, what everybody's been doing in, uh, in all these other modalities where, like, I think a lot of the early successes in, in deep learning have been, like, in images and people are, like, all excited about images and I kind of, like, completely ignored it 'cause, like, you know, an image is worth a thousand words, but it's, like, a million pixels. So, like, the text is, like, a thousand times as dense. So like, kind of big, uh, big, uh, text, uh, text nerd here. But, um, you know, it's very exciting to see it, uh, it take off in, you know, in all these other modalities as well and, you know, tho- those things are gonna be great. It's, uh, like, super useful for, uh, building products that people want to use. Uh, but I think that a lot of the core intelligence is going to come from, from these text models.

   5. SGSarah Guo14:23

      To wrap up our favorite moments from 2023, we have part

8. 14:27 – 17:19

   Arthur Mensch on keeping AI open

   1. SGSarah Guo14:27

      of our conversation with Arthur Mensch, the co-founder and CEO of Mistral, talking about the evolution of collaboration in the AI space and why Mistral's mission is to keep AI open.

   2. AMArthur Mensch14:37

      Models can output any kind of, of text. Uh, and in many cases you don't want it to output any kind of text. So when you build an application, you need to think on the guardrails you need, you want to put on the model output, and potentially also on the input. So you do need to have a system that filters input that are not valid, that you deem illegal, and output that are not valid or that you deem illegal. So, eh, the way you do it, in our mind, is that you do create the modular architecture that the application maker can use, which means you provide the raw model, so the model that hasn't been altered, to ban some of its output space, and then you propose new filters on top of that that can detect, uh, out, the output that, that we don't want. So it can be out of pornography, it can be hateful speech. These things you want to ban when you have a chatbot, for instance, but these things you don't want to ban from the raw model because if you want to use the raw model to do moderation, for instance, you want your model to know about this stuff. So really assuming that the model should be well-behaved is, I think, a wrong assumption. You need to make the assumption that the model should know everything, and then on top of that, have some modules that moderate and guardrail the model. So that's the way we approach it, and it's a way of empowering the application maker in making a well-guarded application. And it's, we think that it's our responsibility to make very good modules that allow guardrailing the model correctly. It's part of the platform. And we think it's, it's the way of, uh, we, there should be a s- some, uh, health- healthy competition on that domain of different startups working on guardrailing the models. And the way you make this healthy competition is not by trusting a couple of companies to do their own safety. It's rather for, it's rather, the, the way you do it is to ask application makers to comply with some rules, so chatbots should not output hateful speech. And so that means that now the application makers need to find a good guardrailing solution, and now you have a competition where you have the, where there's some economic interest in providing the best guardrailing solution. And so that's the, that's the way we think the, the ecosystem should work, and that's the way we position ourselves and that's the way we built the platform with modular, uh, filters and modular mechanisms to control the model well.

   3. SGSarah Guo17:13

      We of course have to mention our chat with the amazing Jensen Huang, co-founder and CEO of NVIDIA.

9. 17:19 – 19:09

   Jensen Huang on how Nvidia decides what to work on

   1. SGSarah Guo17:19

      Here he talks about how NVIDIA decides what use cases to support and what applications of AI he's most excited about personally.

   2. JHJensen Huang17:26

      There are a couple of things that, that our company is shaped, um, and structured to do. There's one part, uh, a very large part of our company is designed to, uh, build very, very complicated computers perfectly.

   3. MSMustafa Suleyman17:39

      Mm-hmm.

   4. JHJensen Huang17:39

      And so that's, that is, um, a, one of its missions. Okay? And, and, uh, that kind of architecture, that kind of organization, uh, i- is a, is a, um, a invention and refinement organization. And then we have, we have, um, uh, a, a whole bunch of, of, um, uh, skunk works, if you will. And the reason for that is because we're trying to invent things 10 years out that we're not exactly sure whether it's gonna work or not. And, and there's a lot of adaptation, a lot of pivoting, and, um, and so, so y- you know, our company actually has, has two different ways of working. One of them is rather organic, shape-shifting all the time. If a particular investment's not working out, we give up on it, move the resources somewhere else. And so that's the agile part of the company, and then there's a part of the company that's not rigid, but it's really refined.

   5. MSMustafa Suleyman18:33

      Mm-hmm.

   6. JHJensen Huang18:34

      And so these two, these two systems have to work side by side.

   7. SGSarah Guo18:38

      Thank you all so much for listening last year. If you want to dive more deeply into any of the conversations you've heard today, we've linked the full episodes in our description. We'll be back next week with more interviews with the leading builders and thinkers in AI and technology. Find us on Twitter @nopriorspod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way, you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.

Episode duration: 19:09