← Future of Life Institute Podcast
Future of Life Institute Podcast
From Research Labs to Product Companies: AI's Transformation (with Parmy Olson)
00:46:37
Loading summary
Transcript47 lines
- [00:00]
A
Ultimately, in AI, this is still a field driven by research and innovation. You know, like the scientists at Google who came together and wrote the attention is all you need paper in 2017 and invented the transformer. Like that was pivotal more than anything Sam Altman has done, or Demis Avis or Elon Musk, like, ultimately, this is driven by lesser known people behind the scenes, the tinkerers and the engineers. There's been a real kind of about face among the leaders in AI today who made these promises to build AI for the benefit of humanity and then ended up pivoting to become much more product oriented. If you're a business, you want to make money, you want to chase profits, you have shareholders, fine. What actually irks me personally is when people try to have it both ways in the way that the leaders of OpenAI do, where they try and speak as if they're still a nonprofit who are doing things for the benefit of humanity, and they're clearly not. When you think of technology bringing us to any kind of step forward towards utopia again, history has never shown us an innovation that does that without some kind of cost to humans.
- [01:19]
B
Palmy Olsen, welcome to the Future of Life Institute podcast.
- [01:22]
A
Thank you. It's great to be here.
- [01:25]
B
Do you want to say something about yourself to start with?
- [01:28]
A
Sure. So I am a technology columnist with Bloomberg. I've been a journalist for, oh, gosh, maybe 20 years. I love my job. I think it's the best job in the world to be a journalist because you get to speak to interesting people every day, learn stories, learn new things, exciting things. And especially in technology, you're talking to people who are literally living in the future, they're building for the future, and you meet some really interesting characters, or I have throughout my career. But I started off as a reporter. I was with Forbes magazine for about 14 years, and then I went to the Wall Street Journal where I was a news reporter. And so coming to Bloomberg was a real change because I'm now a columnist, which means I can write with a little bit more of my own voice and my own opinion. And after coming here and having been writing about artificial intelligence for about seven or eight years, I kind of built on that to, to write this book supremacy. Because when ChatGPT came out, I just thought there was so much more going on behind the scenes that I thought people would be interested to know about.
- [02:38]
B
To what extent do you think the AI industry is driven by personalities and, and, and do you think this is different from other industries?
- [02:46]
A
I suppose it, it's an Interesting question at the end. Is this different to other industries? And I think it is different to other industries because in technology you have this really interesting combination of not only engineers and people wanting to solve kind of logistical problems or technical problems. They have quite kind of societal ambitions to, to, to change society and to make things better for humans. And sometimes these ambitions have even ideological ideas driving them. And I think that creates a recipe for some very charismatic people and characters like Elon Musk and Sam Altman. People whose characters just capture the imagination in ways that perhaps the CEO of a consumer goods company or an oil giant just doesn't. Because these are people who have designs for our very future. And I think that that makes for quite fascinating storytelling and characters.
- [03:53]
B
And it does seem to me that these personalities play an outsized role in the AI industry. On the other hand, it's also just the case that the whole industry is moving ahead thanks to underlying technology trends like compute getting cheaper and data becoming more abundant and so on. Do you think the personalities in AI are what are driving progress or are they riding an underlying trend?
- [04:18]
A
I think it's, you know, more than anything, neither of those things. Actually, I think these personalities are very good at capitalizing on opportunity. You could call that exploiting opportunities, whatever word you want to use. But many of the best technologists are also opportunists. And so I think ultimately in AI, this is still a field driven by research and innovation. You know, like the people, the scientists at Google who came together and wrote the attention is all you need paper in 2017 and invented the transformer, like that was pivotal more than anything Sam Altman has done, or Demis Abbas or Elon Musk, like, this was like a concrete discovery that led to the ability for today's generative AI models to do what they do. And similarly, some of the scientists who worked at OpenAI who came up with some of the mechanics that underpin large language models, you know, that that was like nitty gritty innovation. And I think ultimately the personalities in AI are perhaps the role they play is really driving up excitement and storytelling and hype, and with that comes money. Because I think when, when, for example, when Satya Nadella, the CEO of Microsoft, talks about why he invested in OpenAI that first time, you know, that first billion dollar tranche, it was because he was absolutely enamored with Sam Altman, because, you know, I don't think Satya Nadella was looking at the documentation on GitHub of, you know, what OpenAI researchers were working on or, you know, I think it was his conversations with Sam, where Sam was talking about building artificial general intelligence. This, this ambition to build a kind of godlike software, not just make a better Excel spreadsheet. That's what Satya has talked about, that it really kind of spoke to him. And so I think these personalities play such an important role in keeping this innovation funded in a way, and, and building the story. But ultimately this is driven by lesser known people behind the scenes, the tinkerers and the engineers.
- [06:45]
B
Yeah, yeah, so. So story driven fundraising and personality driven fundraising, how important is that? Because some of the underlying tech can't really work unless you have abundant funding, unless you have enough compute for the transformer to actually provide value, unless you have the resources to gather the data. So is it. There's a framing in which the personalities are fundraising and this is not as important as what the scientists are doing, but there's another framing in which what the scientists are doing can't really work unless you have the funding. So that. Do you see them as working hand in hand or how do you see this?
- [07:24]
A
Yeah, that's probably the best way to put it is a symbiotic relationship. If you have the funding, then you have the resources you need to do the innovation. I was speaking to a scientist at Google recently who was out of the blue, received an email from Mark Zuckerberg with no subject, just. And that's the cool way to do it. If you're Mark Zuckerberg, you don't put anything in the subject line. And he basically was like, you know, we'd love to hire you, please speak to my recruiter. And they were offering this person like triple their salary, just as a start, as a baseline in their negotiations. And in the end, that scientist did not take the job. At Meta, they were a little bit concerned about the kind of political situation over there. They described it as like a Game of Thrones situation. But the reason I bring this up is they were tempted. They were very tempted because Meta has all this money that it's sitting on, this incredible amount of cash on its balance sheet and profits that are coming in from the advertising business that Zuckerberg is funneling into this new division he's called Superintelligence Labs. And what a lot of AI scientists today are motivated by is, is not just money, but the glory of being on the team that builds Superintelligence first or AGI First. And this for them really was a kind of practical cost benefit analysis that they were making. Like, if I join Meta, what is the likelihood the statistical probability that the team I join is going to be the first because they have all this money and all this computer and all this data. So I think, yeah, absolutely, that plays a role. But in the end, for this particular scientist, they made the decision because they didn't think the company was being managed well and the AI division wasn't managed well, that they would not be the first to build AGI.
- [09:26]
B
Yeah, you'll see this in recruiting. When AI companies are trying to recruit top talent, they'll often include information about how many high end chips they have. And this is something that's quite motivating.
- [09:38]
A
It's like catnip for the scientists because.
- [09:41]
B
Then you can run interesting experiments and you can do it at a large scale and you can deploy and so on. We've seen a shift from the AI companies, OpenAI, DeepMind, Anthropic and so on being more like research labs to where they today, I think it's more accurate to call them AI companies. What has happened in that shift? What's involved there?
- [10:06]
A
Well, this was one of the points I wanted to make in my book, which is that there's been a real kind of about face among the leaders in AI today who made these promises to build AI for the benefit of humanity and then ended up pivoting to become much more product oriented. And I think again this comes down to money. Building AI is very expensive and it's also very competitive and there's a real race to be the first to have the best model to, you know, to have the most, the greatest capabilities and the most talent and the most, you know, H100 cluster, the biggest number of H100 clusters, biggest cluster of H100s. And I think that's just been a shift that kind of happened over the last five to 10 years, particularly when it was kind of driven by necessity in a way. OpenAI was running out of money more than five years ago and they just weren't getting it from donations. They had to pivot to become what they called a capped profit company so that they could receive that first billion dollar investment from Microsoft. And from then on it was kind of a Faustian bargain, as it were, that Sam Altman had to make to, you know, if we want to build AGI, we need to kind of compromise on our principles a little bit. Those original founding principles, take the money and use it to eventually build AGI. There's a lot of ends justifies the means thinking, I think by a lot of AI leaders that if we just do this thing, we'll eventually build AGI and then it will all be okay if we build these giant data centers which cost huge amounts of energy. You know, they, they suck in huge amounts of energy, which is not great for the environment at all. That's a price worth paying because AGI will eventually help us figure out how to solve climate change, which I think is. I personally don't agree with that way of thinking. I think it's a little bit. It just makes it easier for people with vested interests to just chase those vents. Vested interests. But that's where we are.
- [12:18]
B
Do you think there's been a change of mission then? Or is it just the case that OpenAI say figured out that they needed much more funding than they originally thought they needed to build AGI and then they had to shift their corporate structure to meet those demands? There's a perspective from which you can say that this isn't really a change in the mission. This is just finding out new information and then pivoting on that information.
- [12:46]
A
So I think it's a change in the mission. And perhaps Sam Altman doesn't realize it. Maybe he can just tell himself over and over again, I'm still building AI for good and I will do that. I think in these companies, people drink a lot of Kool Aid and so do their leaders. People ask me, you know, did people like Sam Altman and Demis Hassabis when they started their companies really want to build AI for good? And I think they did. I think they did start off with quite noble, genuine intentions, but they got kind of caught up in these incredibly powerful systemic forces that are so unusual. Today we have enormous companies like Google and Meta and Amazon and Apple. I mean, they're worth Nvidia trillions of dollars. No company in history has, has been so powerful or so influential. And it's just been very difficult for these two very important AI visionaries who I think they, I think they're the most important AI visionaries. That's why I've tracked them in my book. They sort of got sucked into the gravitational pull of these companies. So when, whenever I hear Sam Altman saying we're still building AI to benefit people, to benefit humanity, I don't believe it. I think the mission has changed. And I'll give you an example. When OpenAI released Sora 2, I think it was last week, this is an app that allows you to create AI generated short videos and you can put yourself in the videos. It's kind of got this meme worthy vibe to it. You can put your Friends in it, Sam Altman released a blog post saying that this could spark a Cambrian explosion of creativity. And when you look at that, when you listen to that from the outset, yes, it could. Because, you know, when you use Sora 2, you really are kind of. It's kind of geared towards you producing content because it's just think about it, so much easier to generate a video by AI than filming one. You know, there's much less mental and physical effort. And so perhaps with SORA too, there will be more people creating. But, but think about it like, is OpenAI really wanted to spark a Cambrian explosion of creativity? Why would they build that on the mechanics of social media? Why do it on the mechanics that kind of underpin these perverse incentives that drive TikTok or reels that keep you scrolling? Because ultimately, this is a product company now. This is not a research lab anymore. This is a business. And you know what? In my mind, that's okay. If you're a business, you want to make money, you want to chase profits, you have shareholders, fine. What actually irks me personally is when people try to have it both ways, in the way that the leaders of OpenAI do, where they try and speak as if they're still a nonprofit who are doing things for the benefit of humanity. And they're clearly not. They've built. They've just come out with a product that is fundamentally designed to keep you scrolling, not to create. Because if it was built to create, it wouldn't be built. Built on top of the fundamental building blocks of a social media app that is designed to be addictive.
- [16:21]
B
And you see the same offering from Meta recently, they also, I think just last week released an AI generated video feed. So this is, in your estimation, this is about earning money and perhaps using some of that money to develop AI or what's the, yeah, what's, what's the most generous framing of this? Because again, you could say, okay, now they're a product company, but they're still doing research. In fact, I, I would, I would guess they're, they're doing more research now than ever, just in terms of raw output, but they're also a product company. And so are they launching these feeds in order to fund their own research?
- [17:00]
A
I think ultimately they need to, you know, they need to make good on the investments of their shareholders. OpenAI just got a $100 billion commitment from Nvidia to help it build out data centers. It has a huge new investor who will want a return, just like Microsoft does, just like SoftBank does. Just like all these other investors do. And so I think that is really what is at the forefront of Sam Altman's mind. He needs all this money to, to build bigger models, and he's making the money, but he also needs to pay back his investors. That's a very important duty. He has a fiduciary duty to his shareholders. Fair enough. But I think that almost certainly will have overshadowed any kind of prioritizing of society and humanity. This is going to be a much bigger deal for him. I thought it was very funny when Sora too got a very, very healthy skepticism from the public and very swift and brutal backlash on social media and X in particular. And people were tweeting things like, well, Sam Altman said that he was going to build AGI to build to cure cancer and solve climate change, but here we are with this product that's just going to keep us addicted to our screenshots, what gives? And Sam actually tweeted a response to that, saying something along the lines of, you know, I get, I get where people, why people are saying this, but we are still trying to build AI for good, but along the way it's nice to build products and technology that people will like. And I think that was just a very, kind of slippery, in a very typical slippery way of, say, of justifying actually where so much of the focus is. It's not just a nice to have, it's a need to have. For OpenAI right now, yeah, yeah.
- [19:04]
B
There's been attempts from the AI companies to tie themselves to the mast, so to speak, to try to retain their original values. For OpenAI, this meant being a nonprofit, then being a limited profit company, philanthropic, this means being a public benefit corporation and so on. And these initiatives haven't always been successful, to put it mildly, I think. What would you change? Do you think there's a better structure or do you think it's just inevitable that companies slowly, over time, drift towards what is most profitable and what makes it the easiest to fundraise?
- [19:44]
A
Maybe I'm just going to be a little bit contrarian here in that I just actually hesitate to blame the companies too much because I think at the end of the day, they are companies. I'm really not a fan of OpenAI starting out as a nonprofit and then making that huge pivot. I think Anthropic kind of has done the best it can to structure itself as a public benefit corporation. These kinds of governance structures that allow you to put the, you know, to prioritize the environment and, and customers and staff Kind of on the same wavelength as, as your shareholders. There's lots of, you know, that's not like a guaranteed way to, to create safe products. But I think the real blame here for me has to lie with regulators. I feel like antitrust regulators in the U.S. for example, have really just not stepped up in the way they should have and they've allowed companies like Google and Meta to grow so, so big and to wield unimaginable power. And I feel this is really like lawmakers and regulators right now just have so little power and there needs to be so much, a much, much greater presence of these kind of overseers, an oversight of AI because Silicon Valley has been left to self regulate for years. And it just, you know, if you're not actually held accountable, you're always going to prioritize growth and profits, which again, you can understand if you're a company. That is your duty to shareholders and to your staff and to your. Yeah, and to your executives. But unless you have the laws and rules in place, you're not going to, you're not going to follow them. So that's where I think there really needs to be change is in stronger regulation.
- [21:46]
B
What role do you see utopian narratives playing here? So a lot of the leaders of the most important AI companies are strongly motivated, it seems, to achieve, to reap the enormous potential benefits of AGI and superintelligence. How does this play out? In their interactions with the public, in their interactions with the government.
- [22:12]
A
So I mean I've, I've, I guess the utopian narrative about AI is, has always fascinated me and what can get sketched out is quite vague. And I think that's, that leads to a real lack of like, I think that's why we have a kind of lack of consensus on what AGI artificial general intelligence will actually look like. It kind of makes me wonder if in the next, you know, few years we'll have companies claiming to have reached AGI and it will be quite hard to refute that because nobody really has agreed on what the definition is, then AGI will become like a marketing term in the same way that artificial intelligence has become one of the most successful marketing terms of all time. I think there's something always very alluring about the utopian vision of AGI. I mean, I love to think about it too. Like imagine actually if AI was just taking care of some of our most taxing jobs and we could all just work one or two days a week and enjoy time with our friends and family and we'd have universal basic income. And we'd be living in this world where these kind of software oracles have taught us how to solve poverty and how to cure cancer, how to find new energy sources so we don't have to use coal anymore. And I just, as much as I love this idea, I just don't see it as realistic and responsible. These are like problems that we as humans have to figure out. And I think this search in particular for a general purpose system that can solve these problems also doesn't make sense to me. I think it makes more sense to focus on narrow. You know, narrow. This is what Max Tech Mark often talks about, you know, this kind of narrow approach to AI and solving single problems rather than this kind of general purpose Swiss army knife of an AI, which actually kind of, I think speaks to the hubris driving this is. Who wants to be the first to build an AI? God doesn't matter, you know, whether it's actually going to solve this specific problem. It's just also the glory that can come with that.
- [24:34]
B
Yeah, yeah. And so you say you don't buy into the utopian narratives. Is that because you, you doubt that AI will become as powerful as it would need to be to deliver utopia or why, why are you skeptical?
- [24:50]
A
I just don't think there's any example in history of people trying to find like a kind of silver bullet innovation that's going to solve all sorts of problems at once. Progress happens really slowly and painfully and it, like I said earlier, it's done behind the scenes by people who are nameless and don't get credit. You know, like people in, in medical history, scientists who will discover something and then they die. And then it's only after they die that their innovation actually is appreciated. You know, that's just, that's been the history of progress till now. It's in fits and starts and it's not chasing one. Again, like kind of silver bullet that's gonna solve everything for us. I think it's a little bit of a fantasy because it just doesn't align with any pattern that we've seen in history.
- [25:45]
B
Yeah, yeah, I can see that. I mean, on the other hand though, I mean, if there was one technology that could serve as like a general solution to a lot of problems, it would be intervening on intelligence itself. That does seem to be sort of the most general technology. If we could replicate what humans can do. Because this is what allows us to solve all of these problems, to tinker and to do science and so on.
- [26:13]
A
Yeah, but in what form or in what guise? I mean, I think there are certainly like innovations, particularly in technology, that have had huge impact, good impact, like the World Wide Web, invented by Tim Berners Lee. He has a book out, by the way, which I'm just like halfway through, which is really interesting about how he came up with the HTML and HTTPs in the CERN laboratory. But of course, if you read his book, he's quite regretful about how the web has turned out because he thought he was going to create this place for. He actually used the word creativity, like this place for creative chaos and just a place where anything could happen and it would create this abundance of great content. But of course, humans being humans, we want to exploit whatever new opportunity there is. And so you have SEO, spam and you have social media kind of addicting people and, and you have, you know, hackers and there's a lot of, actually there's a big price that came with that particular innovation. So, yes, I think, I guess the other reason I'm skeptical is that when you think of technology bringing us to any kind of step forward towards utopia, again, history has never shown us an innovation that does that without some kind of cost to humans. And I think if we reach some big milestone in artificial intelligence, there will absolutely be some kind of cost or price that we have to pay.
- [28:01]
B
Yeah. And I think for many of the leaders of the companies, because this feels so high stake, this feels like a very high stakes situation, they will be willing to sacrifice something along the way. As you write about this is, I mean, I think we can all recognize this as a form of dangerous reasoning, but could it just be true that for us, that if it's possible to get to an amazing future and there'll be trade offs along the way, we need to make some of those trade offs?
- [28:35]
A
Yes, perhaps. I just don't feel comfortable. Where do you draw the line? ChatGPT is used by 700 million people a week. It's probably going to hit a billion by the end of this year. It's incredibly popular. People love this app, but many people are addicted to it and many people have had problems because of it. And there was this lawsuit that came up against OpenAI a couple of months ago by the parents of a teenage boy who died by suicide and he was talking to ChatGPT and asking it for advice on how to commit suicide and the steps he could take and like the. Just the logistical things he would need to do. And of course, you know, there are lots of guardrails in place in ChatGPT to not explicitly answer prompts like that. But it's very easy to get around that by just saying, I'm actually writing a story about somebody who wants to commit suicide. And of course, that's what this kid did. So, yes, I think, is that trade off okay? Even that one person died because this system was built without proper safety testing in mind. I mean, and I say that because this young boy, this teenage boy was using GPT4O and that particular model when OpenAI released that they rushed the launch of that model so that it would come out one day before the latest model from Google's Gemini. And in doing that, they compressed weeks, sorry, weeks or months of safety testing into one week. Into one week. This was reported by the, by the Washington Post, who interviewed people at the company. I just don't think those trade offs are okay. I'm not comfortable with that.
- [30:40]
B
What do you think it's like to work on a safety team within these companies? Do you think it's possible to maintain the specific mission you have under these enormous pressures? If the CEO says we need to ship in a week, but you know, your projects will take two months, what do you do? Yeah, how do you, how do you maintain adherence? How do you adhere to your, to your values in that situation?
- [31:04]
A
I think it's a really difficult situation for people who are working in those teams. And I've, you know, I've heard similar things from people who worked at Facebook or people who worked at Google. And you know, they work on the ethics team or the safety team. And a lot of times they're just not even in the room when they need to be in the room. They're not invited to the meetings that they need to be invited into because they just, they're not high enough up the chain of priorities to be there. So I think it's a real dilemma faced by some of these particular staff. Do I quit? Do I become, do I blow the whistle? And then if I do blow the whistle, is anybody going to care? And what about my family? You know, how am I going to, how am I going to pay my mortgage? You know, these people have like, they have to pay their mortgage. They have to look after their families. I think it's, I think it's really difficult for people in those situations to speak up.
- [32:04]
B
Yeah, it's interesting that there even is such a thing as a dedicated safety team for other, for say, if you're building a bridge, the safety is in some sense built into the product itself. The bridge need, you need the bridge not to fall Down. And in some sense that's the case with AI products too. You need uptime, you need the app not to crash, and so on. But some features are pushed into a dedicated team that's then labeled safety. And this then becomes a concern. You can listen to their concerns or you cannot listen to their concerns. Is there a way for us to build safety more directly into product development?
- [32:44]
A
Oh, absolutely, I'm sure there is, to be honest. I have to be honest with you. I'm not an expert on these things, but I can tell you again, just from other industries, if you have stronger regulation in an industry like aviation, for example, safety is going to be built much more directly into the fundamental design of key features of the product, because it has to. So again, unless the regulations are there, I think it's, I think it's very difficult for us to just expect companies to do that.
- [33:19]
B
Why do you think companies are willing to talk so much about the dangers of the technologies they're building? I think, I think we hear leaders of the companies talking about a bunch about the dangers and they're releasing research showing that their products could be dangerous in such and such situations. Why is that? Because it seems like they could function in another way where they would just say much less to the public.
- [33:49]
A
Yeah, I think it's really interesting. So I think there's in one kind of counterintuitive way when you talk about the danger of your. Okay, so two things. I have two answers to that question. The first is this kind of, I think, counter intuitive approach as a form of marketing that if you talk about how dangerous your AI is going to be, you get this subliminal message across to people that actually your AI is quite powerful and it's a risk worth taking because it can do these amazing things. And it also stirs up excitement and interest in the field. The other. The second thing I would say is I think there's a really interesting phenomenon maybe among tech founders and tech leaders to lean into controversy. And I noticed that with Sam Altman, especially for a long time. I thought this was specific to him, where early on in his career when he made this location tracking mobile service called Looped. This was well before he was even at Y Combinator. He was getting bad stories in the press about his company because people were using it to, you know, abusive husbands were using it to attract their spouses and people were kind of misusing this service. And when he talked to journalists about it, he would talk about the controversy, he would talk about all those problems very, very openly. And I'm seeing that again with other leaders, maybe not all of them, but another example of that is, have you seen this company called Friend? They're an AI startup in San Francisco and they made this pendant that has a microphone on the pendant and it listens to all your conversations. And then it's underpinned by a large language model. You chat to it about your day and about the people you spoke to. They initiated this very expensive marketing campaign in New York City recently where they put posters up all over the subway system showing this pendant. You know, here's your new AI Friend, or you know, much better marketing copy than that, but. And so many of those posters got defaced by graffiti. And what I thought was really interesting was that Avi Schiffman, who's the CEO and founder of this startup, is just reposting all the images of the graffiti, of the, of the, the graffiti posters, and really, again, leaning into the controversy, embracing it, almost posting these images proudly to show almost this. Maybe it kind of goes back to this. No bad news is good news or whatever the saying is in pr. It's, at the end of the day, it's attention. But I just, I can't quite put my finger on what this is, but I feel like there's a phenomenon in tech where a lot of these leaders are somehow able to make the most of criticism and sometimes they even lean into it a little bit and, and it just kind of works for them.
- [37:06]
B
Yeah, I, I can see what you mean here. It also seems like a very advanced form of public messaging. Are they really aiming to. So if, if they're talking about the dangers of the technologies they're building, this, this seems like a risky strategy to, to then hope that the public will get a sense of this is potentially dangerous, therefore it's potentially powerful. You know, investors might think it's, it's powerful, therefore it's profitable. And so I, so I want to invest. Is this, is, is it plausible that their messaging is so advanced and convoluted? Because, like, I could easily see it just backfiring. And if you talk about the dangers, well, then you scare the public. And perhaps you, you can't build what you want to build.
- [37:47]
A
I don't know. I almost feel as though people are becoming numb to the dangers. So, for example, let's look at it on the, let's look at it the other way. Cybersecurity. So I remember I used to write a lot more on cybersecurity about 10 years ago. And I remember back then there was A lot more stories in the tech press about, about hacks and cybersecurity incidents at particular companies and electricity grid hacks. And I think for a while the cybersecurity industry really thrived on fear mongering. And it was very easy to get a story in the press about, you know, some big hack that some cybersecurity company had successfully addressed because people loved to read about this. It was kind of titillating, like the, this, you know, I don't know, someone got scammed or some, some company had all their data stolen or DDoS or whatever.
- [38:47]
B
Yeah.
- [38:47]
A
And then I just noticed this real shift, this kind of shift in the last five, 10 years where we're just not seeing stories like that anymore. And at the same time, I think there's actually been, if you talk to people who work in cybersecurity, they're still complaining about the same old thing, which is that regular companies are not investing in securing their IT systems in the way they should. And I think this narrative of hackers are going to steal all your data. Like, I think the general public just got numb to it.
- [39:22]
B
Yeah.
- [39:23]
A
So I worry that also could happen with AI, that there's been also. I've noticed a real decrease in public, public discourse about bias in AI systems. And I was just talking to an AI scientist who studies bias in systems the other day and I was like, you know, I'm not seeing anyone like kind of writing about this stuff. Is it still a problem? And they said, yeah, it's still a problem. I mean, it's not gone away, it's just very, very difficult to track. But I worry that, you know, these kinds of issues like this and, you know, existential risk from AI, that perhaps the general public is also just getting a little bit tired of hearing about it and they don't care. Maybe they don't care as much because they haven't seen firsthand an example of why that should affect them. Same with, same with privacy. Like, I think for a long time was a big story in the press, in the tech press to talk about the Cambridge Analytica scandal at Facebook and, and then real time bidding for your eyeballs by Google's ad tech systems. But then people don't really have a clear sense of how it affects them concretely in real life and then they stop caring. Yeah, I think that's just how things go.
- [40:41]
B
Yeah, yeah, I think you're right about that. And it's a, it's a, it's a sort of wild effect in the world that we can Lose focus on these very important issues that could affect us personally in the future if they remain issues. We could, you know, we could all be victims to some, some hack or we could all, yeah, we could all be strongly affected by existential risk.
- [41:03]
A
But that's kind of where, why me as a journalist and other journalists like me really have to be accountable because we can't exaggerate the, the potential risks of harms, you know, because if we do, then we lose trust. Yeah, but we, but we also have to spell out very clearly to readers what the risks are so they know. But if we exaggerate to get clicks, you know, and this has just been a chronic problem in our, in my industry, then people don't trust and they don't care.
- [41:41]
B
Yeah, yeah, makes sense. As a final topic here, I want to talk about the ties between AI companies like these, DeepMind and OpenAI and Anthropic and then traditional big tech companies like Amazon and Google, Meta and so on. How do you think the partnerships between big tech companies and AI companies affect what the AI companies are doing?
- [42:03]
A
Well, absolutely, absolutely affects them. I think DeepMind would be a completely different company if it wasn't part of Google as much as the company, as much as DeepMind and Demis Hasabis has said that they are independent, they operate independently. I mean if you just look at DeepMind's website five years ago, even four years ago, let's just say before ChatGPT came out. So three years ago DeepMind's website was all about the research they were doing in healthcare, the research they were doing in energy, alphafold, protein folding. But now if you look at the website, it's completely different. It's all about Gemini, it's much more product oriented. So I think this kind of so called arms race that people like to talk about really has like shifted the focus towards building products rather than kind of pushing the boundaries of AI capabilities and research and problem solving, real world problem solving. And you know, Anthropic, I know they made a real effort to diversify their investments so they've got big investment from Google and they've got big investment from Amazon. But that doesn't mean that there's not going to be a significant influence from one of those investors. I think Amazon in particular Anthropic is largely seen as a proxy for that company's AI efforts. I think OpenAI has tried to also diversify itself away from being so tied to Microsoft and so, you know, they, and Microsoft has done the same. Microsoft has also talked, looked into you know, they've got their own AI team run by Mustafa Suleiman, formerly of DeepMind. They've built their own proprietary models independent of OpenAI. And they've started using AI for anthropic too. And so there's this. I think there's this effort to not be too closely wedded to anyone. But at the end of the day, this is still a very small pool, very small pool of very, very large players. And it's kind of an. And people were saying this a lot when Nvidia invested, pledged the $100 billion into OpenAI. People were talking about this kind of incestuous look to how these companies were all working with one another and money was just pouring from one to the other to help the, you know, Peter, just to help Paul. And, um. And I think that's absolutely true. It's. You could call it a virtuous circle. But ultimately, what does that really mean? Well, it means that you have an established order. You have a few very, very large players with lots of money, and it's near impossible for any smaller player to, to compete because there's this huge wall of financing that nobody else can compete with. Now, we might have a deep seat moment, another one, you know, like we did in January. We're a little company out of China made this model that was as good as ChatGPT for, you know, a much less money. And I would love to see that, I would love to see more of that because I think we're moving into a place where, you know, it's just a few companies are kind of controlling the stakes. A real kind of consolidation of power is happening in the industry and in AI, and unless it gets competitive, then you have just really a handful of people making decisions. And when those, any of those people go, you know, if they have wild ideas about, about the world like Elon Musk, then I think that's dangerous for society. I mean, if you've been on X recently, it's like unbelievably racist and. And there's hundreds of millions of people on this site. So you have one person who has a radicalized, who himself has become radicalized, and he has a particular view of the world and he has shaped a social media platform to support the narrative he wants for the world. Yeah, you can disagree with me and say, well, the previous owners of Twitter had the opposite view. Fine. The problem is through. There's no checks and balances on any of these people. They are just kind of running these systems. Whether it's the AI models that we're plugging into our lives or the social media platforms that we're all addicted to with impunity.
- [46:33]
B
Yeah. Parmi, thanks for chatting with me. It's been great.
- [46:36]
A
Thank you for having me.