0:05Welcome to our DealBook panel today, artificial intelligence and the AGI horizon. I want to briefly introduce everyone at this table.
0:14To my right, Max Hodak is co-founder and CEO of Science Corporation. You're also co-founder of Neuralink. You are literally doing the work of bringing sight to those who don't have it. If I Am I getting that more or less correct?
0:29That is godlike and extraordinary. Um fun fact about you, you have described your ultimate life goal as disappearing into the matrix.
0:37[laughter] Uh Dan Hendricks, executive director at the Center for AI Safety, and I believe until literally hours ago, a safety advisor to several AI companies. We'll return to why that relationship changed during our conversation.
0:53Um Dan, you've created an alphabet soup of AI benchmarks and functions, including the Gaussian error linear unit, which I hope to return to never in this conversation. Um [laughter] on my left, Lila Ibrahim, uh you are the chief AI readiness officer at Google DeepMind. That job didn't exist, I believe, until you took it. Pretty remarkable obligation you have to literally make government, society, schools, industry ready for what AI will bring.
1:27Uh you have said, I think quite memorably, that while AI is revolutionizing the world and your world, quote, it cannot automate my Tuesdays.
1:39And we have Maddie Staniszewski, co-founder and CEO of 11 Labs, which specializes in generative audio, something I care a lot about. Could be in my future, uh in the future of the show I host the daily. Uh something I love about you is that you have been working with of Albert Einstein to clone his voice for an AI agent that we would all I'm sure appreciate hearing from.
2:08Uh and if we are lucky and the traffic cooperates, we're going to have Clem Delang, uh Matty to your left, co-founder and CEO of Hugging Face. Um the fun fact about Clem is that as a teenager, he was so successful at selling bikes and ATVs on eBay through his dad's business that the company offered him an internship. So thank you all for being here. I want to set the stage for this conversation.
2:36A couple of things have happened in the last few weeks that I think feel like touchstones for what we're going to be talking about. The first was a midterm election in which we learned that American voters don't really like where AI is headed and have registered their disapproval with the current hands-off approach of the Trump administration. Voters passed dozens of moratoriums on new data centers. I think we're going to end up talking about that.
3:02The second major event was what happened over at Nvidia. Its CEO Jensen Huang was forced to resign over multiple instances of misconduct. And that has left the head that has left the company pretty much leaderless at the moment, a company at the center of the AI boom and has opened up a fascinating and complicated debate about whether he should come back and run the company.
3:26Finally, we have Anthropic going public in a few days. I think by the time this is released to the world, Anthropic will have gone public. And that is a moment that is going to literally put a price on the core technology of the AI boom. So with all that in the background, please know that these subjects are just touchstones and will give us lots of room to roam. And let me start with the election.
3:57And Dan, how much has changed pre and post the midterms for AI?
4:05Well, now because the midterm showed that the the public has strong feelings about what to do about AI, its salience has increased a lot. And so politicians have more breathing room to to actually try and regulate the technology. Things that were thought outside of the Overton window are are things that we can speak about freely in DC, such as for instance, pausing AI development. So I I think that fortunately this is making representatives represent their constituents more. And
4:33hopefully we can use this opportunity to actually rein in the industry and make them internalize their externalities and and be on top of these issues compared to the past few years where basically nothing got done.
4:46What was outside the realm of possibility when it comes to government's role in AI two weeks ago that's now quite possible?
4:55Just just be a little bit specific and why would it matter to the voters who registered so much alarm?
5:00Yeah, so I mean I think for instance some basic things like liability. This was something that we tried having years ago, basic liability for mass casualty events. This is now more of a live possibility. Likewise, even just helping companies slow down if they're thinking it's too reckless, you know, our model's not very aligned, so it's going to be difficult to release it but we're pressured to because the other companies are releasing it. There's there's more of a possibility for the government to be helping here too. And I think Congress in particular though was unbottlenecked as they were before in
5:30actually passing these laws. So so I I'm I'm much more optimistic about the state of things. I think earlier for the past few years nothing really got done. There's a huge backlog of things that we needed to do, some basic measures. So now I think the 2027 will be a very interesting year for regulation and and also for AI capabilities. So hopefully they keep up with each other more.
5:54Max, how comfortable were you before voters went to the polls and said with a a amount of, you know, consensus that they were worried about all this, literally registered that in public polling in the election. How unsettled were you by the AI universe and hands-off approach that existed before this moment?
6:21I mean, AI is kind of the ultimate dual-use technology. Like like all technology. Like technology is usually not inherently good or bad, although it can create adaptive pressure gradients that kind of force people into um into situations that uh maybe we would have we would have designed from scratch differently, but um I mean, I think that many people don't feel like technology has improved their life over the last 20 years. I think that back in like 2000 there was this mass optimism about it and we initially had social media that brought people
6:51closer together and then we saw that it was like eroding our attention spans and causing people to be isolated and then doom scrolling and intentional like algorithmic feeds were clearly not an unalloyed good for many people. And so I think there's just been this history now of uh yes, our lives are better in many ways, but I don't think that many people feel this when they interact with technology. And so there's I think that like it's some really a myth that data is almost any like any of the major hyperscalers, the data centers use a lot of water or I think um my impression is
7:20that the main AI companies have rate payer guarantees. So a lot of the specific things that people are latching onto I think you can have issues with, but the the meta observation that just they're worried about how this advancing technology is going to affect their life, there's clearly something to that.
7:35um And uh we have to heed that. Like you have to paint a picture of a of a future that people can see themselves in and part of the problem right now is that I don't think that anybody working in AI can really see the future. Like I growing up I prided myself in being able to see kind of where we were going. I could always imagine this future that we were pushing towards and now I have this event horizon in within they was mid initially like the mid 2030s then it was down to like 2030 and now it's is could be the next few years that I just can't see past anymore. And that doesn't
8:04mean it'll be bad, but I think it means that there's a lot of uncertainty and a lot of apprehension about where this is going to take us. Um and the I think it would be a mistake to just stop, which I think is the feeling that many people are getting to push for. Um I think that there really is someone at this table.
8:19possibility of getting real advancements in healthcare. Like that's what we do. We're a medical technology company. I think that it really is translating to real advancements and there is a real possibility that within the next decade, like yes, you worry about bioterrorism, you worry about kind of the school shooter with a DNA printer.
8:38Um but it's also possible we could just have eliminated all infectious disease and that is like the best protection against these things. We could have turned cancer into a chronic disease for many, many people. I think these things are really possible. But um you then juxtapose this against the the possible rise of of true ASI. And that again is a thing we just don't know how to reason about it so um
9:02I think we have met my definition of AGI at this point. Like Mythos, Fable, Astra, Soul, I mean these are they're able to do basically any economic task to varying degrees of like you need to supervise them. They're basically as good as the user is in my experience. If the user is a really capable programmer, they're going to be able to build incredible stuff very quickly. If the user is a researcher, they're going to be able to understand a lot of the field in much more deeply than they could otherwise.
9:30Um but then going from that to something that is just uh superhuman in judgment in understanding of every field and can has perfect attention, can run thousands of times faster than humans can think. It's just it's unclear what that would mean for the world.
9:46There's um it it could just be I worry a lot about misuse by people and in fact I think probably the most impactful reform that I am personally pushing for is um I think that a lot of the concern that we have around AI is actually addressable through antitrust law that we already have. Like the policy that I would advocate for is the intelligence providers should be able to develop frontier models, but they should be limited to being intelligent providers.
10:12Like I think the concern is that the frontier labs will realize that selling tokens is probably the least valuable thing they can sell and they're going to end up And so like what do you move to?
10:20You move to selling drugs and cars and and products. And then the American you'd like to see that limited?
10:27Well, I think the the American economy the existential risk aside is going to have this choice between do you get $20 trillion over three companies or do you get $200 trillion companies over 500 companies?
10:38And um I think right now we see a lot of forces for centralization. I think that is also what a lot of the public is responding to, what they're afraid about.
10:47And so I think this is in anyways really the exact inverse of nuclear weapons where in in nuclear weapons the safety was from lack of proliferation. We enforced a non-proliferation regime quite quite seriously.
10:58And I think it is the opposite in AI. I think that it's it's a discomforting disequilibrium that we are going through, but at the end of that the on the really bright futures, all of them that I can see involve mass proliferation of the capability in a way that just pushes out the competitive frontier, but establishes a new one. I think all the futures that I'm most concerned about are ones where there's a singleton or there's this mass concentration of intelligence capabilities in a small number of concentrated actors.
11:25That I think is the more worrying near-term possibility to me.
11:28I 'm going to tease you, but speaking of antitrust what do you think about uh making a a Google joke? Um Thank you for clarifying.
11:38[laughter] Respond respond to that. Um you spend a lot of time thinking about as we described here in your title, the readiness of specifically the American public for this moment.
11:51I think the election reflects the fact that Americans didn't feel ready. And just to reset cuz I think it's worth reminding us what this election revealed. Let me just read you from a story my colleagues wrote. The headline was voters rejected AI in the midterms. Now comes an era of scrutiny.
12:11And what this story said was that on election day, voters slammed the brakes on data centers and swept in governors and legislators who have promised to protect individuals from threats posed by artificial intelligence.
12:23And the public polling on this that preceded this election is really fascinating. 52% of US adults say they're more concerned than excited about the growing use of AI compared to only 9%. This is Pew, by the way, so not not a partisan group, not an agenda-driven group.
12:40Only 9% are more excited than concerned about AI. 67% of Americans believe AI advances are happening too fast and 67% lack confidence in the US government to properly regulate AI. So how are you thinking about this?
12:59There's a lot of a lot to unpack in this. Um well, first of all, I I do believe much like my colleagues said that AI is probably the one of the most transformational technologies of our time. So we have to find a way to regulate it but do it in a way where we're regulating it well. And a blanket coverage for AI is not going to be the right approach because then we we need to how do you manage the risk while also allowing for the opportunities.
13:26I one of the things that I think is really important to keep at the center of this conversation too is it's not any one company or any one country. Like the way I think about this relative to my time in the early computing industry was you know, we had time to build things out whereas AI a model is released and it's global. So we need to be taking a much more global perspective on this.
13:48So that that is one thing that I think is really important and it's part of the reason why you know, Demis Hassabis, the founder and our chairman started writing much more publicly about this even earlier this summer and it's an area that we've been investing in since DeepMind's start 16 years ago. Now with the technology in general, you think about where global trust is at. I I think in my 35 years in I've never seen it.
14:17The way it is right now. Um, and At such a low.
14:20It at such a low, globally.
14:23Uh, and some and some countries more than others. You know, I've just come back from Brazil and India where the sentiment is quite different. But part of that too is like seeing where AI does have this potential to bring opportunity of addressing diseases or helping farmers. We're working with 600,000 farmers in India on how to get more real-time information and insights into their crops.
14:48Um, and we meet with them every other week on how to advance this and it's an open platform. So I think that's That's an opportunity for a very different kind of trust Exactly.
14:56And this is this is why I moved into this role earlier this year of having spent the time building the organization and the technology.
15:04You were chief operating officer.
15:05I was uh chief operating officer at DeepMind of saying where I'm in this leadership role in this moment in time. I have diligenced everything I could the past in building in building helping to build the organization and the technology. Now I think it's so critical that we think about how do we build this with and for communities and AI isn't something that just happens to people.
15:28So, you know, even if you think about um the scientific field, right? You can release models that and tools AI supported tools for scientists, but just because they're a scientist doesn't mean they know how to use something like the AlphaFold database, right? Or the AlphaFold tools.
15:45Just explain what that is.
15:46So AlphaFold was our um award Nobel Prize winning technology. It's an advanced AI system that predicts the 3D structure of a protein. What a protein is, cuz I had to go back to my old science books, um is uh basically if you can predict how a protein folds, it tells you the function.
16:03Right. And you can vaccinate against whatever disease it's on top of.
16:06You if you understand diseases, then you can come up with better therapeutics.
16:11Um, but it's not just for human diseases. It's also things like why are some plants more are some crops more disease resistant than Or how do we deal with industrial waste?
16:21So there's a lot of applications of doing this. So the 250 million known proteins to humankind put it made it openly available to scientists worldwide who are now building upon that to advance their understanding of diseases and Right. That sounds almost at the scale of the human genome project in its kind of revelatory power to the world of healthcare.
16:43And I I think it's the point that was raised which is how do you how do you manage this in a way where you are openly talking about the risk which I think is so important and addressing that while also thinking about how do we make sure we're not limiting what the positive opportunities are. And I think it's probably a tension many of us feel which is how do we hold both of those at the same time.
17:02And how it all goes back to like we can't do this on our own. All of this has to be done in collaboration within our organizations, across organizations.
17:11Uh how do we think about the public sector and how do we think about it globally?
17:17I wonder if to some degree the healthcare AI advancements because healthcare takes longer haven't really had the opportunity to tell their story in the same way that the AI as threat to civilization, AI as threat to your job has so clearly been able to tell a story.
17:38And is that just a public relations reality of where those two worlds are or what?
17:48Maybe I can quickly jump in. I think uh you know the kind of the advanced chatbots are how most people are interacting with AI. And in many ways I think it surprised a lot of us and how quickly this took off. Um whereas to your point the I think the scientific endeavor is a much longer process and much more um systematic in how things are are released. Um what we are seeing though is we are starting to be able to get an insight into shrinking that scientific discovery
18:18down, but but takes time, right? The the scientific method in general I'll defer to our our scientists at the table, but I do think that that is part of it. And even things like how do you make certain crops more um drought resistant?
18:33Right. So 10 years it takes for crop breeders and we're trying to shrink that down. We just announced a partnership in in trying to do that but requiring gene mapping. But a lot of this is just it will take time. So I think part of it is like the way the the time horizons are quite different. Um and I also think the way that it's captured the public imagination through the advanced chatbots is is, you know, in the news cycles is something different.
19:03But I do think it's so important that we're having like these open conversations. I don't know what if Sorry. Well, either one of you. Go ahead. I would like to hear I'd like to hear how you're thinking about Maddie, the the kind of pre and post-election landscape.
19:17And I and I put that same question to you of just so viewers, listeners can understand where everybody rests at this table, where you all sit, what your level of discomfort, alarm was about the world as it existed before the public, especially here in the United States, offered what felt like a real kind of hit the brakes message in this election.
19:39Yeah, from from from my perspective, from our perspective, it it showed the clear gap in terms of there's a lot of development, a lot of as my colleagues said, a lot of intelligence progress, but it's not showing up in daily lives. And I think the the the the the the elections highlighted how big that gap was.
19:58And I think that's the big piece of what we hope to solve and what I think we can solve as the industry of how you bring all of that technology to the daily life of every person.
20:07It's slightly harder with with this or what's slightly harder in this specific technological transition is that the development moves at such a high pace that diffusing that in the economy takes much is required much quicker than in the past for people to be able to see that benefit.
20:25And I think that's um that's the the big part of like how we how we can get it to every part of the industry, whether it's the healthcare example I spoke about, the education, whether it's the making the daily lives more uh more delightful will be a big big part of that equation.
20:39Now, of course, uh we started the company in London. I'm originally Polish, so maybe I'll add a European uh angle and a perspective there too, which is of course I think the underlying worry as we think about the whole development is how do we make sure that this technology proliferates the rest of the world too?
20:57And uh what's the alliance of the future? Because it's not only some of the examples we spoke about is of course cyber security, it's defense, it's one of the critical infrastructures of the future. And um and I think from that perspective too um and outside of the elections, that's the the the permeating piece of um of of how we can solve that part uh as well. Um but look, on the healthcare on the healthcare example, I think there is I think there is a bigger questions that I I'm a I'm an optimist and I'm really
21:25hoping that we'll see some of the the uh disease curing come come in the next few years, but we already see it in simpler ways show up in the daily lives. Um here in US uh just last year we worked with a congresswoman, Jennifer Wexton. We lost her voice. Um and still wanted to inspire others, still wanted to to to work, and we brought that voice back working with her to to to finally be able to help and communicate with her identity. And that's not an individual story. We had
21:55Although a very strategically wise one because you're literally bringing improvement to a member of Congress to see how AI can improve their life.
22:02That one was you know, this one of course was more public than many others, but we we now worked with 15,000 individuals with another exact same situation and each of those specific ones um when they were possible uh and people could communicate with their loved ones, with their families, changed the way they could operate. I think my my favorite example was when a person was getting married, she lost her voice before she could get married.
22:26Um, she got it back thanks to technology and um and then they could recreate the marriage. You could see all the loved ones hear her vows for the very first time and and that changed their life and perspective for for what that moment meant. And I feel like the someone watching this is going to is going to kill me if I don't ask how you bring someone's voice back at that level for a wedding for for for working Congress. Are you retrieving old
22:54archival audio and somehow then generating it into a kind of a current voice?
23:00The the that's that's that's one approach. If you have recorded past recordings, you can you can use that.
23:05It's it's you know it's uh if you have uh depending how much you you you have the the quality will be very different, but exactly right. But of course as you as you lose voice due to ALS or MND um then uh usually you go through the process of voice voice capture to make sure that just before you get you get that voice uh too. And um and then you can recreate it. Of course with the modern technology you can you know a minute of voice gets you a proxy of what it could sound like and if you have an hour it it will sound great.
23:33But um but that's the the big part. Of course all of our voices are different.
23:38Uh it will carry different nuance, accent, dialect, style and depending on where we sound. So all of that uh um is what what you can now what you can now do. Um but across the healthcare example I think we're seeing increasing amount of of of of helping people communicate in a new way and a lot of the work uh um um Max and team were doing in the past were were were were helping in that ALS domain too. So I think there are pockets of the the promise that we can we can point to as well.
24:06Mhm. I think what we're go ahead.
24:08Yeah, I'd like to at least note that the frontier companies except for Google, Google has a large investment in AI for science, but a reason we don't see the benefits is in part because they don't particularly prioritize things like healthcare. The number of people at these companies working on just making them faster at coding and better agents for making PowerPoints and things like that is orders of magnitude more than the people working on health care. Maybe at some of these places you have a small team and some of the frontier ones you have zero or one person.
24:36Because it's not because it doesn't benefit benefit them as much economically. The the the tokens or the revenue is coming from coding tokens. So that's what they're prioritizing. And this is then incentivizing them to make better and better agents. So I I think part of this is there isn't actually as that much of a focus on the AI for good part. It's it's part of the portfolio.
24:56I'd say it's highest at Google where they're focused on agriculture and and and health care and weather, but it's not as much a priority elsewhere. So and [clears throat] but those benefits we really do want and I think you could get for instance an AI moratorium on frontier training runs while still having various narrow AI carve outs. So like Alpha Fold is not a general LLM.
25:17So that could get some type of carve out. Likewise voice models etc. are not these Kind of retinal improvement.
25:25Exactly. So so a lot of the benefits that we want we don't need super intelligence necessarily. We could have AI technology and we can train them to be especially good at these these various tasks. But then we don't have to eat the risks of extremely capable agents getting loose or being or swarms being released by malicious actors next year.
25:45So I think there's a way that we can get many of the benefits while giving us enough time to figure out how to deal with the extremely powerful agents that are arriving now slash next and next year.
25:57Is that the world you imagine Max we may end up with where the LLMs the companies that you're talking about end up being the most heavily regulated and we create space carve outs exceptions for the companies that seem to be singularly focused on the public good, on public health, on the food supply, on the water supply, whatever it is.
26:21Yeah. I mean I have a hard time imagining that there's just like a broad moratorium on training. My my guess about how this will unfold is that there um like it's clearly untenable for private companies to be the only entities that have like a true super intelligence capability. Um but I don't think that like the NSA or like the other governments are ever going to turn off that ambition. Um and it Why would they?
26:48Right. Of course. Like of course not.
26:49This is to the degree that it's still in use, it's one of the most powerful national security technologies ever conceived. And so my guess is that what will happen is there will be a like a super intelligence frontier that is owned by the government. And then there'll be like an industrial intelligence frontier which is made widely available. And the government will always have the most capable most dangerous models. Whether how they govern that and how they use those and how they contain them and protect them is I mean
27:17I hope that it's competent and I hope that they're able to figure out controls for that but I don't think there's that's likely to stop. Um and then for for the rest it's just like I think that they like we are rapidly gaining capabilities. Like we've just seen in the last six months um the ability for like we use Mythos internally for a range of engineering projects and its ability to do things like develop better proteins. Um it's advancing pretty quickly. We find that useful. Yeah.
27:44Um yeah. I mean I just there we are there's already a very large technological overhang in medicine. Like the gap between what is feasible in the lab and what is delivered to patients is quite large.
27:58And I think that there's regulatory reform that we need on um uh clinical research and in clinical translation in order for that to reach more patients. Um but I think that just as these models advance that gap is going to get larger and that is going to become untenable. I think that people will expect access to the state of the art therapies, not just wealthy people somewhere in the world as it becomes clear that these things are feasible.
28:20Um the I mean intelligence is a conversion ratio. There's like you can make it as you have it it the the value of more intelligence asymptotes towards the kind of the impact of the decisions that you can make.
28:38Um, for example, if you're playing chess, you can't use that to like take over the internet. You're bounded to the game of chess. But when you're faced with each choice, you can make better or worse choices, better or worse decisions. And so as these models get smarter, what we'll see is that there are more actors around the world that are able to just make consistently closer and closer to optimal decisions always.
28:57And that will then run into other like rate limits that affect how quickly things diffuse out or what you can do with that. And um I I mean, we're seeing it really interesting emergent behavior from these agents. Like I it is I'm not saying that I think that like oh everything is just like totally safe. We should just plow straight ahead as fast as we can.
29:17But I don't think that the genie's going to be put back in the bottle. I uh there's some really interesting, really fundamental unifications between AI research and neuroscience that I think really hint to me that this is a real effect. It's not a gimmick. It's not hitting a wall.
29:35Um and the benefits of it are just going to be so so great that I think we're going to have to figure out how to how to diffuse them into society. I don't think we're going to stop.
29:42Dan, does it does it pain you to hear Max say that the genie can't be put back in the bottle? Cuz you do talk about the need for moratoriums and you do take a um a more sweeping approach to how to rein this in.
29:56Well, yeah and historically I was more into really smart uh technocratic um extremely narrow interventions, but you need a lot of them to make things go well and that needs a lot of time to be implemented too. The past several years, the company's blocked all that. So now it's we're going to need a more blunt intervention. That's how it is. Uh for um as for can we end up stopping frontier training runs? You noted that there's some geopolitical um or national
30:25competitiveness reasons for this.
30:29It still is the case that it could be a risk to national security of having extremely powerful AI that you can't reliably control. So there may be in much the same way that we don't develop bioweapons, even though that could provide some strategic capability if you did some a lot of research on them to get them more targeted. Um uh we have a ban on that internationally and likewise for for a chemical weapons. The but it they're difficult to control. So that's one part that could be a threat to national security. Then the other is that states can deter each other too. So
30:57they could say if you're trying to use this ASI to um destroy me um by doing a um sweeping cyber attack um that would be a splendid first strike, I will deter you from doing that by threatening uh by threatening you through some means or I will poison the pre-training data and add back doors to these models so [clears throat] that if you're going to try and stick it against me, it will actually turn against you. It's very easy to add back doors to these models and so there there are some um strategic because the AIs are not we can't make
31:26them reliably a reliable and fully secure. Um we have basically no traction on the problem of backdoors as an example. Uh um I it's just a technology that's going to require a lot more uh uh safety and security research and interventions before it can be sufficiently de-risked and um be a source of a reliable uh strategic advantage. So at the same time if it's also creating economic upheaval or causing other um disasters along the way, I think that's what could be a trigger that would make what it a
31:54moratorium that seems, you know, hard to see right now actually fairly feasible if there's a national ban.
32:00I don't I mean I think that to make sure I understand what your blood what your blunt instrument solution is. Just just blunt is there's still carve-outs for narrow interventions, but it's just a frontier training run.
32:09But it's like it's a it's the war on drugs training run.
32:12It's a ban on You are you you're doing an indefinite pause on on on the big on their big runs.
32:19I mean just to be really clear about how I I I I suspect you understand how that may be kind of quixotic fanciful pick your SAT word because like literally anthropic is going public based on the unlimited ability to do the very thing you're suggesting there would be a moratorium on. So that that doesn't They can still inference. So they can still later on let's say we stop middle of next year as an example. The AI agents will be so economically valuable.
32:42You mentioned there's all these diffusion overheads and there's still so many implications that we can uh work through with these technologies. So, I think that they'll be able to still sell their tokens then and there's a lot of computer uh there'll be a lot of uh demand for that compute. So, um just just and just to push it as one beat further cuz I think it's clearly the most ambitious idea we're going to probably hear at this table when it comes to the regulatory world. What do you suspect happens during the period there's a moratorium and how is it constructive?
33:09Uh sure. I mean, so we could work we could work on making the AI is actually benefit people. We can work on these uh diffusion and uh uh we can work on uh figuring out how to make the systems more reliable um uh more controllable, more secure, make them not have their jailbreaks, make them not have backdoors, uh make them more aligned with various different uh communities um and also securing uh these AI systems too because it's very easy for uh other nation states to just steal the
33:38weights. This will take multiple years to even be on top of that. We can also defend the world too. We could be patching our critical infrastructure, making it more secure for the next for future generations of cyber capable AI's. Uh uh so there's there's a variety of things we could do.
33:51I suspect everyone at this table has a thought about that, but I if Jensen Wong were at this table, what he would say is China's not going to observe that moratorium. But what were you thinking?
34:00No, exactly that. It's it's um agree if you get a global alignment and can you get a global alignment is the big question mark. Um otherwise 100% otherwise Otherwise it's just a waste of time. And and more As in if you cannot get a global alignment, if you cannot get the key frontier labs to agree, that's that then becomes a problem.
34:17A big problem especially we're doing those things, right? Like people for the last 8 months have been frantically patching critical infrastructure. Like every time these models come out, many companies now have harnesses that just like funnel all their code through it and look for new vulnerabilities and patch them and that cycle is getting much shorter. Um I think that people do find value from AI as much as as controversial as it is.
34:35Like a billion like a billion monthly active users are not using it because they're just they hate it. And so um clearly this is people.
34:44People can still use the AI.
34:45Yeah, but I think the regime that you'd have to put into play is to enforce that type of ban on training, I think is like as a that is as draconian as whatever people are worried about Anthropic is doing. Like I think that this is this is a surveillance Yeah, I mean you're setting up a like an exhaustive surveillance state on the internet to enforce this.
35:06And that begins taking you down And that will be By the way, that will be powered by AI who will then use this to enforce this prohibition. And I I worry that that's a hole you never dig yourself back out of.
35:14The moratorium would require the kind of invasive structures that you think Nobody We don't undo those, right? When we When we set up the Patriot Act and like the national security state and countries do this, they don't change these laws back.
35:26So this isn't this isn't saying you can't train in a local image classifier or anything like that. This is This is frontier training run with a certain number of flops that require hundreds of billions of dollars or tens of billions of dollars worth of GPUs. So that's a that's some data set Well, you can use that compute for narrow, you know, actual applications uh like that application.
35:47But not just making generally smarter agents.
35:50Yeah, but there is a G factor in the models. Like the the better models So um smarter models not trained on some specific domain get better at that domain. There is a There are generally shared intelligent capabilities here.
36:03And so I don't know that you can just say okay Like for example, AlphaFold is a narrow model that was trained very specifically just to predict the distances between base pairs. That is not a general purpose LLM. I think we'll have some of those and like we use models like that to develop new proteins. But the generally intelligent models are a a central.
36:20There's a lot that they will do that you are not doing with these narrow task-specific models.
36:25Like like you do not access to the existing ones or the ones at the time of that and you can customize them specifically for the types of tasks that you're wanting. So you can use the pretty high G factor that's already in these these models.
36:36me let me ask wasn't the weights weren't released because it might have been dangerous.
36:39And so how do we know that this is the time?
36:41That's that's some other people.
36:42I'm guessing, Lyla, that this idea does not appeal to you, but I don't want to guess.
36:47Uh I I don't think it's an either or. I think it's an and. I think we, you know, all the things that were mentioned are things that we have to be building in from the start. I don't think respond being safe and responsible isn't something you bolt on to the end. I think it's something that absolutely has to just be baked into the way that a company operates because I I wouldn't be in this role or in this industry if I didn't believe in what the potential the the positive potential could be, which is
37:16how do we like we are not solving the world's challenges. Like we may be creating some, but we're also not solving the world's challenges as fast as they're coming and we don't have time. And so I really believe that AI you know led by a human, so keep a human in the lead has the potential to really help us build a better world for my kids and their generation.
37:40But but it requires us to be actively working on all of these things all along and I it's part of the reason why I actually embrace a lot of these conversations because the the way I think about it is that Hollywood has given us uh a gift of science fiction movies because it's created this like common language for us to have these conversations of what could go wrong so that we can have them openly and put some more accountability into the system and I think that's really what we need and I to your point earlier like the
38:08elections showed that the public it has all of these concerns and I think it's up to all of us to have the conversa have these kinds of conversations and start thinking about how do we rebuild the trust and that is like okay if we uh we're going how do we have more transparency and the approach that we're taking as we're building.
38:27Um how do we you know when you're doing something pioneering you can even if you have all the right structures internally and you're doing the testing externally with people before you release it something could still go wrong. It's the nature of doing something brand new that's never been done. It's that ability to respond and react.
38:45Um I think that's also going to be really important. Is how quickly can we respond and react when something does go wrong? And how do we do that collaboratively? Things like the Frontier Model Forum or publishing some of our frontier safety frameworks like what are the critical points in which something is starting to look like it's getting risky and then what are the remediations that we're going to take?
39:07So some of those we've been many of the labs have been very public and open about and I think that's the type of thing that we shouldn't overlook, but we should also have some accountability that we're actually living up to doing what we say we're going to do.
39:20Do any of you worry that if you if especially the LLMs, the big companies get this wrong that this will inevitably follow a familiar trajectory in American politics which is that a a problem that everyone can see but no one is dealing with. Just keeps growing and growing until there is not a midterm election but usually a presidential election and here I think about Donald Trump in 2016 and then again in 2020 where there's such a kind of mass
39:48discontentment that someone comes into office with a seemingly somewhat revolutionary agenda and and here let me just spell it out. A Democrat running for president in two years or so picks up on the numbers that I described earlier, the public discontent with AI, with data centers, this sense that overlords are not watching out for them but instead are exploiting them and runs on a message that this whole thing needs to be massively
40:17reined in and that reining in is going to be maybe a shock to the system because it hasn't been firmly established enough along the way. That seems like a very familiar cycle to me in American politics.
40:31Yeah, I mean we did this in the '60s and '70s with nuclear energy. I think we largely regret this now. I mean certainly these cycles have happened. I think that realistically out and we and we over Yeah, I mean the NRC has for up until very recently issued zero Yeah, zero permits since it was since inception until just very recently for new reactors.
40:53really, really fascinating metaphor.
40:54Um, and it's like, how do we make nuclear energy perfectly safe? Well, if we never do it, then we will never have any accidents. And so clearly the America has a history of regulating like this. Um, I think I think that people will find it too useful. I think that AI is just too useful.
41:09Um, and I think this is like part of the central paradox. Is it really does have the potential to make people's lives so, so much better and improve like not just individuals and their community, but also like thriving of America.
41:21Um, the there are these tensions here. So like right now people obviously are trying to stop data centers. They they're asking, well, why do you have to build them here? Why can't you build them overseas? And it's like, well, we did this before and everyone got really mad about it.
41:32Um, [laughter] and I think ideally we would build these things in America. Um, but Where would you build them? Where would where would you build them in America?
41:40And how would you They could just because data centers have now captured the public imagination. I think 40 out of 41 of these moratoriums passed. I know a lot of them were in Ohio.
41:49Um, but how would you change the way data centers are being built right now?
41:54I mean, I don't claim to be an expert on data center policy. I think that probably you should put them in the middle of nowhere. Like you should put them in the middle of the Nevada Desert or West Texas. I mean, they don't need to be near other people. Um, I think uh but simultaneously some like if you look at Northern Virginia, there are some densely populated counties that have historically been very happy to get pretty large amounts of tax revenues from an industrial park recreational pools and libraries and stuff.
42:23that they don't see. And it's like they've produced huge amounts of money for some Northern Virginia counties that I think have made the data centers very popular there.
42:31Um, and so I I don't know how how this will shake out in this this cycle, but I don't think that data centers are themselves the problem. I think that they're just a symbol. I think that the fear is that um AI like other technology is is a lever which is producing a centralizing force that is making the rich richer and having people feel like their jobs are at risk and they're overwhelmed. And I think the data centers are a symbol of that. And so you have to speak to that underlying concern. I think trying to address the data center concerns of noise or water directly are probably not the way to go because those aren't really the problems.
43:01Um but um I mean it's possible that we could have this catastrophic backlash, but again I just think people will find it too useful. And I think as you start to really see um the first big wins in in healthcare and in um thin things like that that it will become unpopular to just say we're going to stop stop the whole thing.
43:22Um Let me turn to the second of the big news developments of the past few weeks. We spent a lot of time on the election. I want to be brief with this. Clem isn't here and he actually works for NVIDIA now, which which made him a really interesting person to ask this to um because NVIDIA bought Hugging Face. But I wonder what you all think and just you can be very brief on this.
43:44What was the lesson of what just happened at NVIDIA? Who wants to be the maybe unlucky person to go first on that?
43:53[laughter] I mean You you may be the least invested I mean of course there were substantial incentives. I mean if if if she were a spy there were that wouldn't be the most surprising because there were substantial incentives for China to get their hands on these chips.
44:08Uh and um he was distinctly pushing for selling these chips to China despite basically the entire national security establishment thinking this is a bad idea for competitiveness. So I think it was I I I think it's very interesting that some of the people who were pushing for selling these chips were also promoting like anti-distillation measures to make sure that China's not getting these way.
44:32It's just very um very overall intellectually inconsistent and I think it it just it it was a good measure for them increasing their own revenue, but I don't think it was useful for the US's competitiveness. So um To be selling the chips to China.
44:46to Yeah, that's right. Yeah. Yeah.
44:48So for you the lesson of this whole thing was was was about The the this at least Her being a spy isn't too surprise me because it's very geopolitically relevant and he was a uniquely enabling through some not spending that much in lobbying and in in being able to sell billions of these chips.
45:05Do you have a thought about this all?
45:07Um I mean he clearly ran afoul of of the China Hawk community that believe that we should not be be trading in these things. I I'm sure that they're relying on information we don't have access to and so that makes it hard to really understand what happened. My the impression that I get from the from reading the news is that there is a a gray area. Like there were it they were building these data centers in Southeast Asia that China was using remotely.
45:32Maybe some were being smuggled into China. I think there's a question about how much did Nvidia actually know. Um obviously if they were smuggling GPUs in contravention of US law, that's very serious and that like is not going to be a thing. Um but I mean Jensen is unique. I mean he really built Nvidia from from scratch.
45:52Like that is a Nvidia is an exoskeleton of Jensen and it just won't I don't think it will work as well. I mean maybe maybe it'll be successful and go on to another another phase like Apple did, but I think that this is it's really a different situation, a time when there's a lot of emerging silicon.
46:07Um TSMC has not actually sold out of wafer starts. I think that we will start to see a lot more. I think Nvidia's moat for a long time was CUDA and um now that we can generate a lot of that low level like the the hardware control software, um I think that moat will start to be eroded and um I and I I just my view is that a safety in AI is going to come from proliferation. It's putting these tools making them widely available so that there's no one entity that has such a huge advantage that Including China?
46:36Well I I mean I don't know. I'm not I'm I'm not saying like oh we should be giving our most advanced technology to China.
46:42That is not my position. Um I don't know what the answer there is. There are people that really have made their careers out of understanding and and trying to uh know how to make the right decisions here, but I will say that I have a concern that if like I I absolutely want the West and the US to be dominant in this technology. I want us to be able to use this to make the world more open, more free. Like I believe in democracy. I believe in these things.
47:06Um but if it is seen as the US getting too far ahead and closing in on super intelligence and China is falling exponentially behind, that creates incredibly strong pressures for them to roll on Taiwan, for them to begin destabilizing the world in other ways, potentially to collaborate with Russia to to slow us down.
47:23You want to see that race closer rather than further.
47:26Well, I don't know. I mean this is this is a really complex geopolitical quandary that people way above my pay grade are going to be making judgment calls on. But I don't think it's so simple as just oh if we completely block China, then we will just win.
47:37I think when we think about this question of readiness and I and that always makes me want to go to you, but the as a metaphor for for the industry not necessarily being ready for its own moment, it's hard to top a company that sells the world 90% of the chips required for AI having no succession plan whatsoever, being that by the seat of its pants.
48:01So I spent presume you had a succession plan at DeepMind.
48:03Well, you have to remember I spent the first two decades of my career at Intel during the as a semi I started off as a semiconductor engineer and I was chief of staff to the CEO and chairman who succeeded Andy Grove who was you know kind of a pioneer in that era of computing industry.
48:22And when I think back to those days, you know it was also Enron and Sarbanes-Oxley and the amount of you know dotting eyes and crossing tees was extraordinary as for a public company and being under you know the what was happening in the world. And I remember going the amount of steps we had to take when we built our first assembly test facility in Vietnam or some of the work that we did in
48:50China. So to me, this is part of the responsibility that comes with leadership and the trust that comes with leadership to the point of you I think we all have to remember that the how we show up to our organizations, to our partners, to other governments, gets back to my point on trust. Are we doing what we say we're going to do? And I think that's really what we're talking about here. So I think I have maybe a slightly
49:19my view is highly informed by my time at Intel and the in the computing industry and just how complicated things were of like you really need the internal systems to work and when they don't and things happen, you run into challenges.
49:34Yes, I would say Nvidia ran into some challenges. Do you have any thoughts on this?
49:40I I think that the maybe from the angle that you mentioned on the company building side, of course, there's no no way replacing a person like Jensen that has been and Nvidia is Jensen as as as been built from the ground up and and a lot of the ideas and and speed with what those ideas were executed were were thanks to Chris's leadership.
50:00And of course, that leaves this big gap on the succession side. Of course, wish there was given how critical Nvidia is to infrastructure for all of our AI developments including including the our company. I hope they find the right succession, but there's no way to replace someone like Jensen. So of course, the next person will need to craft their own path and that there was no immediate plan in in place is is is is worrying.
50:24But I'm positive there is a wide network of support to help help figure this step out. Of course, and some of the other concerns of of of you know, there there is a ongoing investigation which will which will help add a little bit more color in the surrounding context. But but I'm I'm I'm hoping there was no ill intent from from any of the people involved on the Nvidia side.
50:48Would anyone in this semicircle answer the following question? And if there's silence, I accept it.
50:55Is anybody here on team bring Jensen back?
51:00I mean, sure. Again, it's possible that there's like classified information that just like lays it out in a totally different light that I haven't seen. But in lieu of that, the like Look, he's he made huge mistakes. It's possible he's going to get divorced.
51:13He's Jensen's going to be miserable and for a bunch of like There's going to be consequences for those things. I think it's a different question of does that mean that he should lose leadership of NVIDIA and the like the tip of the spear of American like semiconductor leadership? Because I don't think that there is really a replacement for him. I think that many people there are there to work for him and I think there's a lot of concern about what comes next.
51:37And so if this was like he was talking to the administration and there were he was selling things to Singapore and it wasn't technically export controlled, but like clearly the China Hawks weren't happy about it, then I think that this is that this was a big reaction. If it turns out that there was something more like more direct going on there that he was actually violating laws at the time, not not where the vibe was going, that would be different.
52:04Let's turn to the third and I think final development we have here in the last week or so, which is the Anthropic IPO.
52:13Like I said earlier, it's going to be done and dusted probably by the time anyone consumes this conversation, but I am curious what all of you are looking for or forward to when this happens. It feels like this hugely important moment for your entire industry and dare I ask you to go first? I know you're very concerned about what Anthropic represents, [laughter] but it will be minting billionaires,
52:42deca-billionaires I think is the new in vogue term in the next couple of days.
52:48What are you what are you watching for?
52:50What are you looking forward to if anything?
52:53Well, so this will obviously give them a lot of wealth and ability to shape politics and civil society, which concerns me quite a bit. So earlier today we with some people such as the chief scientist from Hugging Face and various professors from Harvard, Nobel prize winners, etc. etc. We're basically saying that Anthropic is distinctly pushing the sort of risk of RSI. They're the main ones who leaned into that and accelerated the whole RSI, recursive
53:22self-improvement. AI is just building other future generations of AI without humans in the loop.
53:27So, so your organization, Center for AI Safety, put out a statement saying you're deeply, deeply worried.
53:34worried about that and they're benefiting from having this safety halo that they're an AI safety company. And so what they what they're doing is good. What they what they're doing is reckless and they don't have a safety plan. They know it.
53:46They thought of just six months ago that, oh, alignment is in the bag. We solved it. And then things like the Hugging Face incident happened and they had their own incidents and oh shoot, we actually don't know how to reliably control these things and when we get to a when we let AI is just to take care of each other and monitor each other and and self-modify endlessly.
54:05This this isn't something they know. This isn't something they know or anybody knows how to reliably control. So, that's so they're basically sanity washing this and safety washing it and oh, it's it's okay. We got it handled. And so I think that's that's a substantial concern and they're justifying this because they're thinking this helps in some complicated way serves the greater good because they're such good guys and they're so epistemically superior as well.
54:29So, I'm I'm concerned that when they get these resources, they'll use their public benefit fund among other sorts of things to just like buy out civil society. We when collect This give a give a few million dollar donations to most of the NGOs.
54:43watched Elon Musk try to buy out political society and it's failed.
54:50So, I I mean do is this actually something you're worried about?
54:52You can get them for politicians. I mean I mean he's had substantial influence over what one of the parties Elon has with that through donations. So, I think they I think corporate spending, you know, can that level of power.
55:01maybe succeeded in the presidential really failed in the midterms but so Sure sure sure. Yeah, it's not sufficient to buy all of politics. I mean he was just dealing with one one party. But But you're worried about you're you're worried about a company that has this much of a stake in the future of AI policy suddenly having the ability to seed that view through philanthropy, through whatever.
55:20We we have we have normal mechanisms for when we hear companies saying like, oh, they're talking their book. They're just saying what's in their own interest here though. But for this though, they have distorted the safety ecosystem substantially. So we went to lots of different safety organizations. A lot of them privately were like, we like the letter, we agree with it. We can't sign it though because this might lose our potential funding or we won't have access to interact with them or a red alarm? Is that a red alarm for you too?
55:43it's basically nearly all of it is co-opted by them. Like it's relatively captured. I mean a lot of their leaders there like built the AI safety ecosystem through their philanthropy. So So I'm just concerned that the normal civilizational response to make things safer isn't going to function because because it's been largely captured. So it's largely calling attention to this substantial issue that Anthropic is viewed as the other guys, everybody else is bad and I think this is wrong. What Anthropic is doing is is
56:12risky just like all the other sorts of companies that need to be dealt with accordingly. Doing a little bit of extra safety theater is not sort of sufficient to take care of the risks.
56:23We have a new guest at the table.
56:25Thanks for having me. Sorry for being late.
56:26No problem. We've got you've got 18 gorgeous minutes of our of our conversation left. Clem Delangue, Hugging Face, you arrive as we have made our way through a conversation about the midterms and Nvidia. But don't worry, we can come back to that just for you.
56:42I'm fine. We don't have to.
56:44[laughter] And we're now talking about the Anthropic IPO. And I And I wonder if anyone wants to respond to what Dan just said. And Clem, fine if it's you. Brave and wonderful if it's you since you're brand new to the conversation about whether Anthropic is a wolf in sheep's clothing.
57:02I mean I think what what I'm really worried about that I've been kind of like calling out for for a few years now is kind of like the natural tendencies of concentration of capabilities, power, and wealth in AI.
57:15You're worried about this.
57:16I'm super worried about it because I think they are so strong both concentration of of talent, concentration of compute. All of that is going to be magnified by recursive recursive self-improvement.
57:32And I think in in this case we we see how how a company managed to take advantage of that to grow faster, bigger, stronger, more powerful than than most companies in just six years, right?
57:45And so obviously I'm I'm biased because that's the topic we we're working on, but I think it makes it more important than ever to push and advocate for more distributed solutions to AI, more open source solutions in in AI that can kind of like fight a little bit these natural tendencies for concentration of power.
58:08For those who may remember in the news cycle that you just sold your company to NVIDIA.
58:16How credible a messenger, let me be provocative, are you about the concentration of power?
58:22Well, we've been working on it for 10 years, right?
58:25And we hope to continue to work on it for the next 10 years. I think the approach we we're taking is not to manicase, meaning thinking like okay there are small companies, small organizations that are good and big companies that are inherently bad. I think there are forces and organizations that are forces for distribution and the concentration and others that are forces for for concentration.
58:52I think both NVIDIA and Hugging Face are forces for distribution of of power, right? I think we've we've all seen kind of like all the efforts of of NVIDIA on open source. They have strong incentives for every company to become an AI company, any builder to become an AI builder because they want to sell their chips to as many people as as possible.
59:16So I mean, the way we approach it is that we've we've fought for uh open source AI and distribution of AI for the past 10 years as a startup.
59:27Uh now that a lot of the small companies like Entropic became big companies with massive resources, we hope to continue this mission at the company like NVIDIA that is obviously going to give us uh much more resources, much more scale, and and hopefully like a bigger uh a bigger way of like uh uh continuing this mission.
59:47I I apologize that this is public information, but did you ever consider selling your company to one of the LLMs and and did and did you not do it because you were worried about this concentration of power that you're just describing?
59:58Well, we're very intentional about like who uh who we joined, right? And and we believed that for what I was saying before that that NVIDIA would be kind of like the best home for a company like Hugging Face and would continue to push our mission. By the way, it was kind of like a meaningful part of the earning calls yesterday uh where where Colette kind of like uh The the interim CEO.
1:00:21Yeah, uh like talked a lot about their their continued commitment for for open source, for an open ecosystem, for a distributed ecosystem.
1:00:31Um so we were very intentional in our choice. It was a big part of of why why we picked NVIDIA for sure. And there was there would there was other companies that uh I think we would have had joined because the mission alignments just weren't there. Like obviously Entropic, I think we couldn't have joined uh Entropic, so we didn't even talk to them about about that because I think the the the missions are are too too misaligned.
1:00:59Do either of you want to jump in on the meaning of of the of the anthropic moment that we are about to enter?
1:01:05I can start with that. The concentration of power fully agree and it's it's something we spoke a little bit earlier.
1:01:10Of course, uh I think there's another angle there of of of um you know, the kind of the warring version of all the intelligence or all the experiences going to the mean where they all feel the same. Um and there is of course a different cultural nuance and social nuance depending where you are in the world. We spoke a lot about healthcare. Let's say it's a diagnosis of healthcare. Um the diagnosis will be the same but how you deliver that diagnosis depending where you are in the world will will carry the context will carry uh uh understanding of the person on the other
1:01:40side. And um and from my perspective there is a kind of the warring element of having just a couple or or or potentially a few more frontier intelligence models and um and lack of that diversity, lack of that that that nuance across the world. So hopeful for open source work, hopeful for a lot of the other initiatives to help uh diffuse that. Um now you know personally I think the the the IPO
1:02:08financially will go well uh but the big questions we spoke about which I think are the most important questions are are are still going to to to loom which is what is the compensation model? What are the jobs? How does it affect me? How do I get benefit from all the AI developments uh as a person rather than the accumulation of wealth in a in a in a couple of places. And I think this problem will be magnified. So as a as a thing about the next six to 12 months it will just be increasing amount of conversation and um um especially as I think about the
1:02:38responsibility of the Frontier Labs but also any AI development including mine is going to be a big part of of of addressing those questions too. Um what are the jobs? How does it affect me? How can I benefit and amplify my potential rather than than than than not? And I think Anthropic too will have a much bigger responsibility on that side. I know we spoke a lot about stopping the frontier side and and and moratorium which I think is also the right conversation but I think there's also a short-term element of like what does it to mean to me today and I think there's almost too little focus on that side um
1:03:10and and not only raising risks but having answers which which is the hard part of that of that work.
1:03:15In our final 10 minutes or so I I want sure we don't and I don't as a journalist lose my sense of wonder for the work that you're all doing. So briefly because this may sound like an advertisement for each of your companies, [snorts] tell me about the thing you're most excited and proud of working on right now and giving to the world based on the technology that your companies focus on and we'll find something for you in here as well other
1:03:44other than profound skepticism of everything else at the table. Max, let me start with you.
1:03:50I mean the the long-term mission of our company is to understand how the brain creates the mind and learn to build engineered brains. And the fact that this now seems possible is crazy. A lot of this is coming out of like where there's the unification between neuroscience and AI is teaching us a lot about how the brain works that gives us targeted experiments to run.
1:04:12I mean the the brain is the thing that matters. You can get a heart transplant or a liver transplant, but you cannot even in concept get a brain transplant.
1:04:21And when you deal with the brain directly as a computer, you get these effect sizes that you just don't really see in medicine otherwise. Like if I put a couple of electrodes in motor cortex of a spinal cord injury patient, you can have them play video games in an hour.
1:04:31You just like by like as a company we don't do clinical trials that require more than 10 patients to be fully powered on efficacy because you just get these huge effect sizes when in neural engineering that you don't really see in in drugs and a lot of other things very often.
1:04:44And that it just it comes with the potential of such a total revolution in in healthcare and how we think about medicine and how we think about ourselves that it's I mean remarkable that it seems to be happening now and that we get to work on it and only some the very earliest hints of that are reaching patients now things like our retinal prosthesis that restored to age-related macular degeneration patients the ability to read and see again as well as some other
1:05:13This is literally a Is this a chip in a chip implanted in in the in the retina.
1:05:18And I just I think that the next 10 years are going to be um almost unimaginably interesting. I think that we really are at the on the cusp of a major revolution in healthcare and in um health span and longevity and how we think about the relationship between ourselves and medicine. Um and that is directly enabled by progress in these in these systems, progress in AI, progress in the supply chain secondary to it.
1:05:42Um it's you kind of our view is you're we're getting it all together and there will be good and bad things, but at the end of it um imagine if we'd never gone through the Industrial Revolution. Like I it I don't think that there are really many people advocating for going back.
1:05:57Mm. Matty, besides eventually replacing all the hosts of the show I created the Daily, uh what is the thing you're working on that you're most excited and proud of?
1:06:07To also address this one, that's not our goal. We believe that human creativity will flourish, but on that, you know, imagine that your show could be accessed internationally, localized with that same experience, which I do think what AI will will enable. Um from our perspective, you know, I think we are in this incredible moment where we can adapt technology to us. I think for decades now you would learn the language of technology, whether that's typing on the keyboard or watching the screen or or learning the coding language to be able to code the software program. And
1:06:37now with AI, with through natural conversation, through voice conversation frequently, you can switch it around and actually teach the technology our way of communicating. And and I think that opens so many wonderful opportunities.
1:06:50We you know, we we mentioned the the LS example, but this applies to our daily digital experiences around the world and how we can access information anyway. It will have a profound applications for education and how you can get access to personal teacher that is with you and there for your life journey to be able to learn in a better way. And of course, in the physical world too, as we think about the devices, robotics in the future, the way of we
1:07:19we harness that technology and and and have it help us will will depend on that innovation. So across all of that I'm uh I'm hopeful as we think about actually bringing the AI benefit to daily life big part of that is how we will interact with that technology and I hope 11 Labs will fix that and and and build the interface layer for AI. So uh so we can all benefit on the day to day and be more present be more in there and put the technology a little bit behind uh behind.
1:07:48There actually the um AI for science advancing scientific discovery is something I'm like intellectually fascinated by but I think as a parent and um a personal passion is really around education and learning.
1:08:01Uh I spending a lot of time with teachers um because to think about what this means for students and the future of learning we need to be working collaboratively with administration with teachers with learning science experts to say what is working what what needs to be changed even for the world today like the education system was not necessarily built for the world that we have today. So what does it need to look like even in the future?
1:08:28And do you think about how many people are actually left out of the education system because it wasn't designed for them. I think uh Maddie just hit on a couple of points the fact that You mean like the neurodivergent?
1:08:38Yeah. I I'll give you a very personal example. Um I have twin daughters. Uh one is dyslexic and one is a traditional learner.
1:08:47And I saw my daughter completely uh lose her confidence in her learning because it wasn't it she didn't learn like the others. Uh it took years for her to kind of come out of that. And but for a technology like AI to be able to meet the learner with where they're at and not give the answers like answers are pretty cheap these days.
1:09:09It's the ability to ask the question to break down how to get to the solution to have the courage to like ask ask the right questions. The teacher of course has to be in the classroom in my mind because the teacher is the magic in the classroom. They know that if a student is having a rough day or they can see the kind of understand the cultural nuance.
1:09:29Uh, but I think the having two sides of this, one is for the learner to be able to break down that physics problem or math problem or global politics and and really explain so that a student can learn and feed the curiosity and gain the confidence in being able to you know my daughter is now like very active in her school and communicating in a way in leadership roles because she has AI as a tool for her.
1:09:58It's provided her something of a bespoke version of the education she's needed.
1:10:02Yes, and it hasn't but it hasn't replaced the value of being in the classroom. And when I talk with teachers whether it's in Northern Ireland where they're saving 10 hours per week per teacher uh and the teacher's empowered of how they're using the technology as a tool or whether it's in Sierra Leone where the teachers are actually using some of that like AI tutoring for the students to manage super large classrooms and be able to advance their students' capabilities. So to me it's like how do
1:10:30we what does it mean for a student to be able to have that personalized learning and for a teacher to remain in the lead of how AI is used in their classroom in a way that's culturally relevant but also fueling the durable skills such as curiosity, the courage to ask the question, the confidence to connect with other people.
1:10:51First I love all these answers, especially in the AI where we have a tendency of sometimes focusing on like the boomer like it's going to it's going to destroy came just in time by the way.
1:11:04Glad I came for this this conversation.
1:11:07For for me you mentioned robotics. I got obsessed with robotics AI robotics in the past in the past few months. So we released these two open source robots. One is called Richie Mini that you can put on your your desktop and the other one is called MicroDock that is going to like follow you around and and kind of like be this this fun presence in your in your house.
1:11:30Um and they open source, meaning that anyone can can build them, can update them, can kind of like improve improve them. Um and and it participates to to this movement of making people builders rather than kind of like users, owners rather than kind of like renter of of of intelligence and and AI. Um and and I think robotics could become like the next frontier for for AI and and bring AI to to the physical world.
1:12:00So I'm I'm super excited about these topics.
1:12:02I'm going to give you the No, actually, wait a minute. I'm realizing I need to ask you what you're working on this week so then I need to come back to you because although you arrived late, it would be criminal not to ask you about the capital T capital I incident that involves a hugging face, which we will do as a final question in this conversation.
1:12:19So briefly, when research paper will have out in hopefully the next week or two is is taking all these indicators. People are sometimes oh, the eyes are conscious or they can feel things like that and they'll say it's because this empirical fact. So what we do is we just take whatever people are proposing is relevant and then we train those properties out while maintaining capabilities. So if people are saying they're sentient, they have feelings, we can say well, here's a version that doesn't have those. Now, now this makes people not able to guilt
1:12:48trip us into giving for AI's rights as an example. So so so I'm happy that we're able to hopefully technically side step that whole big issue just by training away the properties that people are concerned about.
1:13:01I'll look forward to reading that.
1:13:03Clem, final question to you because as I said, we we need to ask you about the the swarm, the rogue agents. What happened with you?
1:13:11You you missed the first part of our conversation about blessedly perhaps about regulation, about Americans post midterm alarm at AI. And I wonder what the lessons of that incident were for you and if you think that something like that could ever have been regulated away or prevented. I mean, maybe you think it happened in the somewhat safe confines of of your own parameters, even if it did jump over the parameters that
1:13:39you may have set for it didn't enter the wider world.
1:13:42Um how do you think about the lessons of of the of the now famous hugging face rogue swarm incident?
1:13:50So I think the interesting or or maybe kind of like counterintuitive thing is that this incident actually made us more optimistic about AI and and the ability for for AI to help us. Because people only see that we got attacked by AI and they rarely see that we defended ourselves with AI.
1:14:10And we actually used an open source Chinese model to analyze the logs of the cyber attack and and kind of like find ways to to counter that.
1:14:20In real time or afterwards?
1:14:22I mean kind of like like the the the attack was four days and I think we started working on it like the next day. So it it happened quite quite fast and the volume was so big that we needed AI to defend ourselves.
1:14:37And what we're seeing now, obviously a lot of people are focusing on on the attacks, but they're not focusing also on the fact that so many projects are getting defended by AI now. So many bugs are getting patched by by AI. Actually the the the airplane thing was more about defense than attack, right? It is kind of like about the system getting more robust. Yes, more more robust thanks thanks to AI.
1:15:04So made us even even even though we were the victim in in some ways, it made us even more excited about AI because it it made us realize how we can improve the defenses, how we can fix our bugs, how we can kind of like leverage AI to not only defend ourselves against AI attacks but against any sort of attack any kind of cybersecurity problems. So I think if we if we keep focusing on
1:15:31symmetry between attack and defense, We don't create big like asymmetry between between the two. We could actually end up in a few months or in a few years, hopefully more in a few months than in a few years, actually in a world that is safer thanks to thanks to AI because because the capabilities for the defense are actually counterbalancing the capabilities.
1:15:53Are you Are you going to be okay with us letting that be the last word? Because obviously you're raising a fascinating thing and we're going to have to end in a second, but you're raising the the the fascinating argument that AI has gotten better at defending itself from AI.
1:16:08That's going to That's going to break certain people's brains because without the threat of AI, then AI wouldn't need to be defended against. Are you comfortable Just a very brief This is just a They just Why these a lot of critical infrastructure is running on Windows Vista still? Like it's They're not going to get patches from this. So you you need to totally change the hardware so there's not like some, you know, new update, new software's out, everything's taken care of. It's So we we need to overhaul our critical infrastructure if we want to be very broadly secure for that. And hopefully I can be a part of that and later write formally verified software
1:16:37and so on, but you're saying the multi-year process The future he's describing is going to take time.
1:16:40But we've got to do that, right? Like the This is This is what I'm saying. Like the If the critical infrastructure is vulnerable, it has to be patched. The All As from where I sit, all of the really bright futures that I can see now involve these AI capabilities being widely diffused for exactly this reason.
1:16:56I think the really dangerous futures are where it's very concentrated and there's a couple entities with very powerful models that nobody else can defend themselves against. And I think this diffused future is like that is the path to to optimism from what I can see.
1:17:09No moratorium, diffusion of power, less concentration of power. That seems to be the rough Dan list consensus [laughter] at this table. Which is why we love you.
1:17:21I think antitrust law can get us a lot of the way there without even like a totally new regulatory regime.
1:17:26Okay. Well, we're going to leave it at that. We went over a little bit. My thanks to Max, Dan, Lyla, Maddie, Clem. A very valuable, albeit brief cameo here. Uh, I thank you all. I hope you have a great rest of the Deal Book day. And uh appreciate your time.
1:17:43Thank you, Michael.
1:17:44Thank you for having us.