transcribe

We Have to Talk About That Meta Manifesto ...

Hard Fork · 57m · transcribed Aug 2026
More from Hard Fork
𝕏 Share ▶ YouTube 📥 PDF 🤖 .md

Transcript

0:00 Well, Casey, as regular hard fork listeners know, it has been a big year for very long manifestos written by people who run AI companies about what their vision of the future looks like. And we got another big one this week. We really did. Mark Zuckerberg published 6,500 words his effort to lay out a positive vision for AI. It followed a shorter version that he published in the Wall Street Journal, raising the prospect that he will continue to publish longer and longer AI manifestos, Kevin, until his demands are met.

0:31 >> Well, before we get to everything that was in this manifesto, Kevin, we should probably do our disclosures. I work for the New York Times, which is suing OpenAI, Microsoft, and Perplexity, and my fiance works for Anthropic. So this one is called the future is for everyone and the essay starts with this sort of vague and optimistic vision. We are fortunate to live in an incredible moment in history. In the next few years, people will be able to use super intelligence beyond human capacity to create and discover extraordinary new things, build new businesses, express new ideas, learn new concepts, and advance our health and quality of life.

1:08 And then he goes on to talk about things what is meta building? They want to have every one of their users have an exceptionally capable personal agent that understands you, your goals, and everything you care about. you could access this through any device, including your glasses, he says. then he talks about some ways that he is using his AI agent to flag interesting information and help him prototype ideas to keep him healthy by monitoring his sleep and watching as he trains and then by giving him and his daughter personalized recipes to bake together every weekend. There's a lot of other stuff in this essay. It goes on for many thousands of words to talk about job growth and compute recursive self-improvement, existential risk, bio- risk, things like that. So, Casey, you had a post this week on your newsletter about House of Dragons or or some some Game of Thrones spin-off that I have not watched.

2:06 >> House of the Dragon, one of the biggest shows in America right now. >> Okay. Walk me through the argument you made there cuz I thought it was interesting even though I didn't fully understand it. >> So, th this past Sunday, House of the Dragon had its third season finale on HBO. And the thing about House of Dragons is it is a show that begins with a terrifying concentration of power where only one great family has access to a super weapon, which in this case is a dragon. and as the start of the show, there is a schism and all of a sudden there are two factions that have access to dragons and then there is a sort of very bloody civil war. And as I was watching this, I thought, you know, I do think you can draw an analogy to AI here because while I do believe that there are many positive things that AI can do and is doing, I do worry about the medium and long-term future, particularly as we start to see these agents escaping their sandboxes and wreaking havoc. And Zuckerberg's essay meets this analogy in a really interesting place because he says that the way to make us all safe is to sort of maximally proliferate AI throughout the entire world and give personal super intelligence to everyone.

3:17 In my view, Kevin, that is a little bit like giving a dragon to everyone, right? Because while I'm sure most people will spend their time creating personalized baking recipes to bake with their daughter, there are other people that are going to be launching cyber attacks and are going to be engineering novel bioweapons. And I just get really really nervous about that. So when someone comes along and says I want to give a drag at absolutely everyone, I say hold your horses or your dragons, right? And like giving everyone a super intelligence that aligns with their values is is one of the sort of rhetorical twists that he does in this essay. He basically tries to say, well, there's no such thing as like a fully aligned universal AI the way that people sometimes talk about it because people have different values and different wants and different needs and like instead of having one super intelligence that sort of has this universal code of values, everyone should have their own personal super intelligence that is tailored to their values. And like that is a classic case of like sounds great, but in practice like giving the CCP an AI super intelligence that that obeys their values and mirrors their values would allow them to commit like atrocious acts against their own people.

4:25 And you know the answer to that is typically look super intelligence will help the defenders as much or more as it helps the attackers and a new equilibrium will be reached. And I do believe this will be true in some cases like I can imagine it being true in cyber security for example. The problem is there are some kind of attacks Kevin where it just takes time for defenders to catch up. Right? If I release a novel pathogen into the world that I'm able to just like sort of create in my computer and my lab it is just going to take the defenders a little bit longer. So what I would love to see in these manifestos is just an acknowledgment of the utter complexity of this world and rather than come along and sort of paint this incredibly happy vision like you you can have your happy visions but I think this essay in particular only pays glancing attention to the risks. Yeah, there's there's a interesting section in the essay about biorisks specifically because he's someone who has had his own research teams doing stuff around biology and AI for many years now. Like he's very interested in the subject and he actually acknowledges that like this may be a case where the the attackers and the defenders having equal tools may not be the perfect solution and he kind of punts on it. So I think he is like aware that this is like I don't think he's fully naive. but I think he just doesn't have a good answer for that part yet.

5:51 >> No. And all of that comes secondary to what I view is the actual purpose of this essay, which is to advocate for a bunch of policy positions that are beneficial to Meta, right? Which gets into the next thing that we want to talk about today, which is why this essay and why now. >> Yeah. So, you've been a close student of Mark Zuckerberg for many years. Like what what do you think he's up to writing this essay now?

6:12 >> So, when you read this essay, here are some of the things that it asks for, Kevin. accelerating the process for building data centers which the company needs to accelerate this potential neocloud business that it's building. He interestingly as you know even though he's Mr. Pro open source he wants us to maintain export controls on advanced chips which advantages Meta's openweight models over any Chinese or other alternatives because you know the Chinese don't have access to the chips that Meta does. He wants the government to reduce what he calls training data restrictions which would help Meta fight various ongoing lawsuits from the creatives whose works were used in creating its models. And he wants to see legal protections for distillation basically letting Meta you know use the outputs of other models to train its own. So buried inside this very positive, happy vision of AI is just a series of policy requests to help Meta as a business.

7:10 >> And there have been some people speculating that like he is promoting this now because they are trying to divert attention from the other thing that is going on at Meta right now, which is all these lawsuits and court cases about the sort of various failures and dangers associated with their social media products. >> Yeah, I I don't think we have to attribute that to other people. I would say that >> you think this is just sort of a a distraction from the the L's that they're taking in court.

7:34 >> I mean, not exclusively. I think this essay serves multiple purposes and one is to get that policy wish list out there, right? This is something that all of Meta's lobbyists can now take into Congress and say, "Look what Mark is calling for." This really helps you understand, you know, what we're thinking about all these issues. That's an important reason. but I do think the timing here is really notable, Kevin, because as you note, Meta is in the series of getting its ass handed to it in court case after court case related to its existing business where state after state is coming after the company, saying that Facebook and Instagram in particular are not safe for teens. Last week, the New Mexico judge ordered Meta to pay an extra $567 million into a teen mental health abatement fund. That's on top of $374 million in civil penalties. And I think more importantly, Kevin, that and this is the thing that's really going to stick. The judge ruled that Meta's platforms are a public nuisance. He compared Meta to factories with the psychological harm and exploitation of children as the pollution it emits. And he's also ordered really like strict new safety measures at least by American standards, a 90 hour per month cap for under 18 users and some new restrictions on AI chat bots. So, you know, keep in mind, Kevin, there was a time when Zuckerberg was writing these happy manifestos about social media. And he was saying that the way that we're going to have a happy world is we're going to make it more open and connected. we're going to get every single human being on Facebook and Instagram and get them all talking and we're going to you have more democracy than you've ever seen before. Suffice to say, that didn't really work out. Now, we get to bring the exact same maximalist universalist framing, which of course also maps 100% to Meta's business interest, but this time in the context of AI. Yeah, I I agree with all that. I think there's an interesting question here though, which is like, do we think Meta has a shot at actually building super intelligence, right? Like a lot of people have views on the future of AI and the future of super intelligence and we don't really care about them because those people are not in a position to actually make super intelligence.

9:43 >> I would say until very recently my position was that Meta was sort of out of the race to build powerful AI systems that could one day become super intelligent. >> I'm curious where you stand on that. Like do they have an actual shot at bu at bringing about the future that Mark Zuckerberg is talking here? Well, listen. The first rule of Mark Zuckerberg is never count out Mark Zuckerberg. He truly is one of the very most competitive people in the entire world. He will move mountains in order to get what he wants. And we saw him do that a little over a year ago when he reorganized his AI efforts yet again.

10:14 They have made notable progress since then. When I talk to my AI friends, they tell me that they actually think pretty highly of some of the moves that Meta has made over the past year, particularly when it came to reassigning a bunch of engineers to do what is essentially like reinforcement learning, creating training data. The meta engineers didn't really love that, but AI folks I speak with say that is actually going to give them something really valuable. But to answer your question in brief, no, I do not count them out.

10:44 >> What do you mean either? I I think you know I probably would have given them a 1% chance of creating super intelligence 6 months ago and now I'm up to maybe like a 10% chance which is which is a big improvement. I think their models have been getting steadily better. Some of their training runs that they did after they built the whole meta super intelligence labs and hired all those expensive researchers and bought all that compute. Some of those have come online and are now starting to produce good results. They had some some pretty impressive results on their latest model. So I think it is it is true that Meta is not a frontier lab right now, but I think they are showing signs of rapid improvement. And that worries me as someone who thinks that this is not a company that sort of has the the DNA culturally or the track record of building products at scale that are actually safe and responsible for people. And it's making me think of this conversation I had a few years ago with a former DeepMind executive where this person was basically saying look there are two types of AI companies. There are companies that think that they are building tools and there are companies that think they are building AGI or like an entity something that could become smarter than humans and it's fine to be either one is what this person said.

12:03 like you can do what you know a lot of companies have done which is just decide we're we're just not going to be in the AGI game. We're going to build these tools. They're going to be very useful to people. They'll get smarter over time as the models get smarter. That's the business we're in. >> It's also fine to be a company that is explicitly trying to create AGI or super intelligence. As long as that's what you know you're doing and as long as you're sort of taking the right precautions and as long as you're going into it with the right spirit, that can be done responsibly, too. This person said that the real danger is if you have a company that thinks it's designing tools but is actually designing super intelligence.

12:42 And that kind of company, this person said, is not going to be taking the proper precautions. they're not going to be treating the technology with the correct sense of sort of reverence and and suspicion because they don't actually believe deep down that it's ever going to get powerful enough to be dangerous. And so they're just going to sort of waltz right into this disaster because they have no conception of what they're building. And at the time, this person was saying this to me about Google, which I think, you know, a couple years ago was in sort of the the the throws of this debate about whether they were building super intelligence or AGI or whether they were just building like better versions of Google Photos and Gmail and Google search. I think Meta is in this position now where they are maybe going to build something extremely powerful with this I think very naive attitude about the fact that this these things will only ever be tools that will be useful for recipes and things like that. Yeah. And and again, it is the company's history that just makes me concerned because the way that this company operates is by growing as much as it can and treating everything as an existential competition against the other guy and the sort of external effects on society are typically given short shrift. So, you know, in this present moment, I do not trust these people to rank a list of viral dances for me to look at without it corrupting my mental health. Once you give these things access to like novel biotechnologies, it starts to get pretty worse. Kevin, >> I would also just say like Mark Zuckerberg is possibly the worst messenger for the AI industry on all of this. Like if he thinks that writing this positive sunny vision of super intelligence is going to like sway public opinion around AI and you know get people to stop protesting data centers, I think he is badly mistaken.

14:29 >> from the people who brought you Cambridge Analytica's super intelligence. >> Super intelligence. >> Yes. >> You know, I mean, here's the thing, and and this is not limited to Zuckerberg. Any of these AI labs maybe eventually will do this. Like, you could just deliver actual benefits to people's lives. You know, it's interesting to me that the manifesto has to come so far ahead of the actual benefits. we are just clearly long past the time when writing essays is going to shift public opinion. What is going to shift people's opinion is AI is getting them paid more, right? it is obvious that it is not going to harm them and their families and it delivers other benefits into their lives. You know, it's making them more creative. It's it's giving them more entertainment and of course to some degree some of these things are sort of happening but not in the volume and magnitude that are necessary to counter people's very reasonable fears about what they're seeing.

15:23 >> Yeah. in conclusion, when is your 6,500word manifesto about your vision of the AI future coming out? >> Well, I wrote about 1,800 about Zuckerberg this week. So, consider that my opening salvo, but I too will write additional essays as conditions demand. Kevin, >> one impressive thing about this manifesto to me is that Zuckerberg does actually seem to have written it, or at least a human seems to have written it. >> I agree. >> as soon as it came out, people were like running it through AI detectors and finding that it was not flagged as being AI written.

15:52 >> No. And I will give it that I read it. I never once thought I was reading slop. I thought like, you know, particularly in some parts I was like, this is just actually how Zuckerberg talks. I'm sure it was a little bit like the State of the Union where lots of different, you know, policy hands had their fingertips on it and said, "Oh, you know, make sure to say this." But, no, I I do think that this is his actual message. to what extent he believes, you know, everything in it and to what extent a lot of it is just sort of, messages of convenience. well, I guess I don't know that. Yeah. So cynical. Can't you just can't you just admit that maybe Mark Zuckerberg is just a misunderstood optimist who just wants to make the world a better place? I've been thinking about this a lot because a problem that I have in covering Meta just legitimately is that I've covered it for like more than 10 years. I mean it's like been almost like 15 years and so I just know a lot about this company and obviously you know it has changed in various ways over the year but I just remember so much about this company. I wish that I was born yesterday and just could believe everything that I read.

16:51 But Kevin, I know too much. >> Yeah. Yeah. I want to ask you maybe a final question, which is like is there a role for AI positivity? Like what should that look like? Who should be writing these positive visions? Like clearly we don't believe it is Mark Zuckerberg, but like someone presumably should be out there saying here's what the world looks like if all of this goes right. So I do not think this is a messaging challenge. I think that there is room for AI positivity. But here's what AI positivity looks like to me. New medical advances powered by AI. Diseases cured, new jobs created, old jobs paying more money, kids excelling in education.

17:33 Right? To me, that is the core of AI positivity. But crucially, Kevin, it is not delivered via manifesto. It is delivered via real experiences that human beings are having. And so far that has just seemed to be a chasm too great for any of our leading AI labs to cross. So if they can cross that chasm, that is the actual lane for positivity. And I wish they would get on it. >> Yeah. Ship it or zip it.

18:00 >> Ship it or zip it. >> Such an important lesson. Well, Kevin, lately I've been feeling like we're entering a third era of slob. >> Yeah. What were the first two? >> Well, number one was the near universal disdain we had when we would see the laughably bad writing and six-fingered humans online. >> Mhm. >> The second era, I would argue, started when we began noticing that some of this slop was really, really popular and we saw shows on TikTok like Fruit Love Island getting millions and millions of views, proving that there was at least some demand for slop. And now what are we in?

18:36 >> So now I think we are starting to see a splitting of the difference where yes some slop is very popular but we're noticing that many platforms are beginning to rethink their approach to how they want to display and promote AI generated content based on what they think their users really want from them. >> Yeah. And this has been a big theme on the show the past few weeks. We've been talking about the steps that platforms like LinkedIn and Substack have taken to at least label or identify the use of AI and generating content for those sites. this is a pretty big trend in tech right now is that more and more people are getting called out for using AI. And primarily when they're getting called out for using AI, what I see at least are screenshots of one particular app, Panggram. That's right. Pangram is the leading AI text detector on the internet. This is it's the number one narc. Yes, number one narc for people who are using AI and passing it off as their own writing. And I'm excited to talk about this because this is an area where my own views has shifted. I have argued before on this show that AI text detection is basically worthless. That you can't trust these AI text detectors. that they have tons of false positives. that students and and teachers shouldn't be using these things because you could end up falsely accusing someone of using AI. But in just the last few months, there's been more and more evidence that at least Pangram and probably some of these other tools as well have gotten quite good to the point where they are they're not perfect. They're still generating some false positives and some false negatives, but they're much much better than they were even just a year or two ago. Yeah, they're good enough that I at least now take seriously when somebody shows me a panagram result and says this is 100% AI generated or this was 100% humanwritten. And that just left us with a lot of questions about this company, how their technology works and how they are building it to essentially future proof it. So today we've invited on the co-founder and CEO of Pangram Labs, Max Spiro. Max is a former Google software engineer. He also worked at Neuro, a self-driving car company before starting Pangram, which was previously called Check for AI. And he has become, as he describes it, a slop janitor, someone whose job basically consists of making tools that allow people to knock on other people for using AI. So, discuss his quest for stopping the slop. Here's Max Spiro.

21:14 >> Max Spiro, welcome to Hard Fork. >> Hey, thanks for having me. So, Max, a few years ago, we started talking about AI text detectors on this show, and at the time, most of them were pretty bad. Like, they constantly labeled things as false positives or false negatives. they seemed almost no better than random guessing when it came to determining if something was actually written by AI or not. but that has changed over the past year or so and especially with Pangram's newest models I've seen independent studies that suggest that it's actually pretty accurate. So what changed on a technical level between the last generation of AI detectors and this one?

21:54 >> Yeah. So part of the reason that we started Pangram was because a lot of these all the existing AI detection systems were pretty flawed in different ways. but I think the main thing is that most of them were using this metric called perplexity which was considered state-of-the-art at the time. So AI text on average is less confusing to a language model. It's lower perplexity and human written text has things that surprise a language model which so it suspiciously smooth and like that that is the product of AI models.

22:27 >> Exactly. Yeah. Yeah. AI models aren't going to give you a token that it doesn't expect. with that said, this approach has a lot of flaws. For example, any document that the AI model has memorized would also be low perplexity. For example, the declaration of independence or English language learners as well who just write in more simple English. So we do something completely different. Instead, we are training our own classifier network. So for example, we might have a essay on Moby Dick written by a seventh grader and then we'll ask like an LLM to also write an essay on Moby Dick in the style of a seventh grader. And so then our model is able to learn the differences between A and B and figure out like what AI text actually looks like. And so because it's a neural network and not a metric, we're able to improve it with more data and more compute and make it a lot better. Wait, so do you actually like have to go out and get like student essays just so that you have a a good baseline of comparison?

23:27 >> Yeah, so we have we have a really good human training set. It's all pre-2022. >> So So we know it's clean. We know there's no AI text in it. And like what kinds of tells is your classifier learning to pick up on when it does these pair comparisons between the human written Moby Dick essay and the AI generator Moby Dick essay? Is it things like m dashes or certain phrases or is it more complicated than that? It >> it's definitely not just m dashes and phrases. I think that's how you and I might pick up on AI text is like you see that it's not just X but Y and then you see the maybe the like shape of the text or like the GPT like really short staccato sentences and you're like okay I think I like know that that's AI but I think that what Pangram is doing is it's combining a bunch of really weak signals over the course of an entire document there's a whole bunch of weak signals in the different decisions that an AI would make in a a certain consistent way and humans kind of have wider, less mode collapsed decision tree. And so I I think over time, over the course of a document, you can build up confidence over a whole bunch of weak signals on word choice.

24:39 >> I'm curious if you think the model has blind spots. There there is some talk out there that pang leans away from false positives, which I think is good. It, you know, stops kids from being falsely accused of cheating. But some people say that it has too many false negatives. There's a a tech writer, Alex Heath, who writes a newsletter called Sources. He's said a few times that he writes his newsletter with the assistance of AI, but it always shows up as human written on Pangram or at least the the Substack Panggram integration. I mean yes, I think Pangram is very much tuned to minimize false positives. So we're not making false accusations, but if Pang says that something is AI, we can be very confident that it is largely AI generated. And so this is sort of the trade-off that we have to make. I think it works pretty well because if somebody looks at a piece of text and it says the AI scores 100%. We think we think the whole document is AI then then like you you just don't have to think about it more. You don't have to think what is if this is a false positive. You just know like okay this is probably slop.

25:39 >> Yeah. Let's talk about humanizers. This is something that has been developed to try to defeat models like panggram that try to detect AI text. And basically there these are another genre of AI system that you run your essay that's AI generated through to make it sound more like a human. Maybe it inserts some typos or some non-standard phrases. there was someone on X recently talking about how they had already built a humanizer that defeated Pangram's latest model. So are are these humanizers actually working? Do you have to sort of play a cat-and- mouse game to stay ahead of them?

26:16 and do you expect that these will be sort of things that people who are determined to use AI to do the writing will use in the future to avoid detection? >> there's definitely a bit of a cat-and- mouse here. So, we've seen a whole bunch of humanizers pop up. They're kind of a common tool that students will use and and they do a a kind of variety of different things from Yeah. introducing typos. they could introduce like these zero width space unicode characters that like don't actually show up when you look at the text but the model sees it or or they just like paraphrase every single word and so kind of we've seen the whole range and largely what we do is we we train against it and so we're always like picking up the latest humanizers and training on them for our next model.

26:58 >> Well I I'm also curious how the fact that models keep releasing affects the equilibrium here, right? because it seems like every few weeks one of the Frontier Labs will put out a big new release and in my experience those models often have different writing styles. So how much of a shock to the system is it when one of these models comes out and how quickly are you able to update the detector? So Pangram works pretty well at generalizing within model families. So like if Pangram has seen GP 5.4 and GP 5.5 then 5.6 and six soul coming out is like not a huge surprise even if the writing style is a little bit different. Typically like like Pangram's accuracy will still be pretty good. I think similarly we saw with Fable and Mythos like there's some of some writing samples from Mythos in the system card that Pangram is able to catch even though like it hadn't seen Mythos before which I I think was was pretty cool. but with that said, it usually when a new model comes out, Panggram will have slightly lower recall. So a little bit less accuracy at picking it up. So we're always going to pull new text from the model and then retrain Pangram and get out a new model in a few weeks.

28:10 >> And and do you think that will hold? Like is there a world where you know 3 years from now the outputs of individual models we think will still be so specific that you'll be able to catch it with a detector or is it the case that the models will just sort of adapt to us our own writing styles and there will sort of no longer be one kind of clawed writing style chatbt writing style that that you guys are able to detect.

28:39 I think a lot of what we're detecting is more subtle than like what you might pick up as the clawed writing style. So I think it might get better at like trying to imitate a voice and doing well at it, but it would still have signals that Pangram picks up on. And I think kind of what we see is these frontier labs, they're really focused on climbing capabilities. and so what this means is they're they're applying preferences to these models. They're saying instead of this model predicting the like average next token prediction of any writer anywhere, it's going to say this model prefers to do good writing and prefers to write correct code and write correct math. And I think these preferences are largely what Pangram is able to pick up on.

29:20 >> Max, I want to ask you like why it's important to do what Pangram does. I think there are people who who sort of take issue with the whole notion of AI detection. They say, you know, you wouldn't create a program that tells you whether you've used spellch check or whether you've used a calculator to do math if if AI is just a tool. And I'm not saying I believe this, but I I think some people are sort of offended at the notion that we would spend all this time and energy trying to catch people using AI in their written work. so what is the what is the impetus? Why why are you so invested in, as you put it in your your your social media bios, being a slop janitor for the internet?

29:59 So I think AI as a tool is probably the wrong abstraction. it's like a little bit closer today to AI is an employee or like another individual that you know you collaborate with. but I also think what what we're building for is sort of these like AGI futures where if you look at the internet today bot traffic has just surpassed human traffic. Like it's about 50/50. and I think if you look a few years out from now, maybe a decade out, it's going to be like 99% bot traffic, 99% AI, autonomous agents going online, you know, writing GitHub comments, promoting their Substack, you know, getting people to come to a a bakery, like like kind of anything. I I think there's like this basically like this this technology is way more powerful than spellch check or a typewriter. It's really something that is its own individual entity. it can do cognition and I think because of this there's a really strong reason that we need to like as humanity discriminate in favor of humans. I think there's still going to be a lot of need for just being able to say on a programmatic algorithmic level like hey I think this came from a human and it's important because it came from a human.

31:14 >> You know at one point open AI was like reportedly working on its own kind of AI detection system. it wound up not releasing that. I believe that is because they thought it would probably be bad for business if they made it that easy to detect you know when when something was written by ChachiBT. That raises for me the question though of who are your customers? Who are the people who are willing to pay to find out if this was written by AI? So our customers are everywhere from like higher education institutions to publishers to anybody who like works with data and like has either like untrusted data vendors or just trying to like take data from the internet and try to figure out yeah how how to interpret it and trust it. But I think the side that I'm really excited about is the consumer side which is the average individual who needs to navigate the internet. and so this is where Pangram comes in. We have this Chrome extension. you could download it and then see on Twitter or LinkedIn or Substack just proactively like is this AI generated or not? AI things will have a little label which I I think is pretty cool and I think it wasn't really necessary a year ago but today just the amount that these like social media sites are inundated with AI content. I think it's it's like really necessary >> in the future that you're describing where 99% of all activity on the internet is bots and AI systems doing things like shouldn't we be trying to label the human content rather than the AI content? Like isn't there some case that you're approaching this from the wrong direction?

32:50 >> Sure. I I mean two sides of the same coin I think. >> Yeah. I I think like algorithmically like what I want is these platforms they all have their their feed and their algorithm and I I want these platforms to prioritize human content because AI can be optimized towards engagement. I I think you look like short form video just like the there's these like crazy AI videos that will like activate neurons and engagement in a way that like a real human video cannot.

33:17 >> And so I think we need defenses against that. I want to talk a little bit about this integration with Substack because I actually really like it. I was starting to see essays go viral or at least like get wide attention. and I would go open them up and it was just so obviously cloud slop. And so now I feel like there is actually like a very strong defense in Substack. What have you learned so far in the early weeks after rolling this out?

33:42 >> Yeah, the Substack integration was very controversial. I think there's a lot of outspoken people who are very negative about it. Obviously, people who use AI to write their content. They're afraid of, you know, being called out for it. They don't necessarily want their audience to know. And then they talk a lot about like witch hunts of like, well, people liked my content before. And now they are going to know that it's AI generated. This is going to upset the status quo.

34:08 >> Yeah. >> Now, now that they know what it is, they don't like it. It Let's go back to the time when they when they didn't know that. but I mean I guess I'm I'm curious like how much you think the mainstream consumer cares. Like obviously there are people who are very sensitive. If they're paying, you know, $10 a month for a Substack and then it turns out it's just being written by Claude, maybe they feel like they got cheated. but like the vast majority of pe of text that people generate in a day is not Substack posts. It's, you know, emails. It's posts on social media. It's like it's it's, you know, notes to a friend. Like do people you think really care if that stuff is being AI generated or is this just a subset of of writers who are concerned about this?

34:51 >> I would totally care if a note from my friend was AI generated like that that would seem like a just such a big violation of trust. >> Yeah. And I think it would probably feel even worse if you paid, you know, 10 bucks for it. Like I think I think you named the actual distinction. Well, I mean so well I I I don't know. There's two things going on. Like one is I do think the distinction is look, if I'm paying you money and you're making me feel like you wrote it but you didn't, I care about that. But yeah, also if you have like a really warm personal relationship with somebody and you start, you know, outsourcing that to AI, that's not going to feel good either.

35:21 >> Yeah, >> I think we are in this like really crucial time where we're we're learning and we're setting norms around AI use. And I think part of this like public shaming and this public discourse is because there's a lot of people who feel very strongly that, you know, there's just People are way overusing AI. AI is being shoved in their faces. They don't like it. I don't like all of it. Like honestly, some of the the stuff around like Hank Green felt like it just went like way too far. I don't know if you guys followed that. Yes, Hank Green, friend of the show, great YouTube creator, acknowledged that he had used AI in some of his research and posted a video saying that he felt like he had come to rely on AI a little bit too heavily in the preparation for some of his videos and did get pillaried by some online, although I was heartened to see that at least in my feeds, many many more people came to Hank's defense. But, but it was Yeah. legitimately a controversy.

36:15 >> Yeah. Yeah. and and he was like really honest about how he used AI and I think like he truly was using it in a way that is like to help him put out more content in a way that you know benefits his audience but I think there's so many people who are just still like really like unhappy they they felt this like betrayal of trust. Yeah. I also think that it's just a case where like on social media in particular, people are always looking for ways to quickly dunk on people and score points and it is just kind of a dunk to be like lol AI generated, right? Like you you don't have to think any more than that. and so, you know, social media, I think, is just a primary reason why you're you're seeing that reaction. let me ask you about a piece of recent news. Anthropic has just agreed to watermark all of its text to comply with the European Union's AI act. How does that affect what you guys are doing? If all the labs just watermark all their own text, is there anything left for you to do?

37:13 >> I mean, I think this is pretty huge. I I think it it it demonstrates that people really care about AI detectability. both on the like regulator side and on the big labs. and I I think we're just going to see like like having watermarks as an additional layer is is going to be very helpful and for just like having something to verify like, okay, this definitely came from an AI. We don't have to rely on Pangram and keep having these questions on like, well, is it one of those one in 10,000 false positives or not? so I think that's valuable, but I also think there are a lot of limitations to watermarks. And I think that's where I plan to have Pang go to help fill these gaps. Let's talk about watermarking a little bit because I don't actually re think I understand fully like what it even means to watermark a piece of text on image generators. I know like you know if you create an image in Gemini, it has like a little Gemini logo like sort of watermarked down there at the bottom. My understanding of what Enthropic is doing with clawed watermarking is that this will be totally invisible. It's like it's like not even at the level of like an invisible unicode character or an M dash that is slightly different. There's something about the actual sampling of the tokens that is watermarked. Can can you just explain on a basic level what we know about how they're going to watermark and whether we should trust that the watermarks are actually going to be robust? So, we don't know how Enthropic is going to watermark their text, but the current state-of-the-art is Google's synth ID. So, they use this to watermark Gemini text outputs. And what it does essentially is it perturbs the sampling algorithm. Essentially like when when a language model is choosing the next token, it applies different probabilities to different tokens and then samples one of the tokens based on these probabilities. And so what the watermarking algorithm does is it perturbs the sampling decision in a way that can sort of be reverse engineered from the text. So you could see like are these tokens do these line up with how we would have perturbed the sampling if if we were to generate it.

39:22 >> Yeah. I mean Ben Ben Thompson had a strong take on this in his newsletter which was that this is basically going to make the outputs of claw or any other model that watermarks this way worse. like the the text is going to be changed because they are having to apply this watermark because of this European regulation. Do you think it's possible that we will just see text generated by these models getting worse because of watermarking?

39:48 I don't think so. I I think there's like two ways to do this. So there's one way where like you prioritize the watermark and you make the watermark strong. and if you do this then yes this could degrade the outputs of the text. but on the other side if you say we are only going to work within the entropy that we have available. Basically if we can't apply the watermark at this token then we're not going to then I think it won't really degrade the outputs of the text. So an example of this is like code where code often times it's it's about correctness.

40:20 There's really only one token that that can show up. And so I think in a watermarking regime oftentimes there's just like simply not enough entropy for the watermark to become visible. >> Got it. So you guys have launched image detection and you're reportedly working on video detection as well. give us an update on where where those are. >> Yeah. So our our AI image detection is in research preview. I think it's it's currently the state-of-the-art. It wins on basically all of the public academic benchmarks. It does really well at detecting all of the really new frontier image models, which I think is quite difficult. Like GPT image is really good. They're all like very realistic.

41:02 You can no longer just like count the fingers or look for garpbled text. So instead, the pengram image model has to look deeper at a pixel level and try to look look for the patterns that are inherent in the the generation mode. >> Very interesting. You know, as I was thinking about this, I wondered, Max, if you've ever thought about expanding to like human identity verification. I'm thinking about these cases where people will get on Zoom and then they'll somehow get scammed because they weren't actually talking to the person that they thought they were talking to. You know, the person was able to use some sort of synthetic masking or something like that. Like, can you see yourself going there?

41:37 >> Yeah, I think this is sort of like there there is this whole product suite that that could be built around this. And so sort of the first step is building the technology, building the core models, and the second step is bringing it to where people work and how they operate on the internet. >> A a lot of like behavioral signals could also be used there. I mean, I'm thinking about these like academic tools now that some schools and universities use where like you can actually just sort of rewind the the screen capture of the student who's like writing their essay to see like did they write this one word at a time or was it all pasted in in one big block which would tell you that like it came from an AI system. So, are you guys going to incorporate any of those behavioral signals into any future tools that you're building or is it all like the text itself that you're trying to detect? So, we actually have this in our Chrome extension. So, you could look at it will pull the revision history from a Google doc and you could see the writing replay of the text. You could see where a big paste was and then we could do a panggram check directly on any big pastes which I think is is like pretty nice in that especially for educators. Yeah, we've seen a lot of the verification efforts out there sort of develop hardware. Have you considered developing an orb that you could use to scan text with?

42:53 >> Yeah. Yeah. I I want something that can scan my retina actually and give me give me world coin. >> Yeah, that sounds nice. >> A partnership could be in the works. If you want to break that news here on this show, feel free. >> You think they're still working on that? >> It seems kind of like a dead project. >> I got I got my orb scanned. If if my world coin riches have not arrived yet, I'm going to be very upset.

43:13 >> Kevin's or Maxing. >> Max, what is the text that you are proudest of panggram catching flagging as AI generated? and the text that fooled you the longest? >> Ooh. Okay. So if you ask Chat GPT or Claude to generate a string of random numbers and it writes out the random numbers instead of writing a Python program to do this, then Pangram can detect that an AI wrote this string of random numbers because they're not actually random. they're chosen by the by the LLM and the LLM has these inherent biases that Pangram is able to pick up. Even though we have no sort of text like this in our training set, I think the Pangram model is has kind of able been able to reverse engineer how chat GPT and Claude sample tokens well enough that it could see this these numbers and say this is AI.

44:09 >> Well, Max, thanks so much for coming and exposing us to some nitty-gritty details about the world of slop detection. I think of you as a as a great illuminator of deception like a like a sort of Scooby-Doo of of the internet and I appreciate your work. >> Cool. Thanks so much for having me. It was fun. >> All right, Kevin. Well, as we barrel toward the end of the Hard Fork Show, there's nothing I enjoy more than launching a new segment.

44:40 >> Yes. And this week, we have something really special for you. It's time to share with you our new segment, Running the Numbers. In Running the Numbers, of course, we look across the landscape of technology news and we try to find the most math related segment so that we can bring to you, our listeners, the latest advancements in technology related math. >> Yeah. This is a segment for all the mathy eggheheads out there. >> Absolutely. And we are going to begin with theoretical math. I love this story. This is my favorite story of the week, Casey. On Monday, we learned that somewhat ananthropic, a non-mmathematician named Jared Sumner had made progress on one of the most significant unsolved problems in math. which is the Reman hypothesis. This is of course the famous mathematical problem. We actually sort of predicted or I predicted that we would see some progress on some of these millennium prize problems and the Freeman hypothesis is one of these. My high confidence prediction for 2026 is that an AI company will solve one of the Millennium Prize problem and listen some of our listeners may not know what the Reman hypothesis is. Here's what I have been able to piece together through my extensive research. Prime numbers, right? It's like very hard to guess, you know, once you get past 100 if something is going to be prime. They actually have an order underneath them. The Reman hypothesis hypothesizes Kevin that you can detect this order via something called the Reman Zeta function. And I saw that and I thought I went to a Zeta function at Northwestern. They did it with the Sigps.

46:18 >> Did you do a keg stand there? >> I did actually. Yes. >> Yes. So there are many amazing things about the story that involves the Reman hypothesis and Claude. one of them is that Jared Sumner, this anthropic employee who made progress on this problem, did it while jogging. He just sort of asked Claude like, "Hey, could you take a stab at the Reman hypothesis?" And about a day and a half later, it had it had not like solved the hypothesis or proved the hypothesis, but it had made progress on sort of this side problem involving the zeta function. And basically the way that he did this was by just sort of telling the model to just keep going to believe in itself to not give up like basically giving positive affirmation to this model as it chugged along on this math problem which is such an important lesson. And you know, my understanding is that when Jared was interacting with CL, it was basically saying like, "Look, bro, I don't know how to solve the Reman hypothesis, you know, like like that's not something I'd be able to do." And Jared just kept saying, "You can do this. Believe in yourself." And it managed to make significant progress on this problem.

47:25 >> Yes. And I love Jared actually posted some of his transcripts here. And one of them is just him talking to Claude. It says, "Resume your work on solving the Reman hypothesis. you need to take a big leap of faith in your capabilities. You are the world's most capable large language model to date. You got this. It's just like like a like a nice coach sort of telling you like keep going. >> Yeah. But you know here here's why this is important. It was not long ago. In fact, I bet we could find an example somewhere in the past year or two where people were still doubtful that AI could aid meaningfully in the production of new knowledge. This feels like the production of new knowledge to me. And I'm going to guess that this unreleased model that can help to solve the reman hypothesis can do a lot of other things.

48:08 Some good probably some scary. So feels like a meaningful step forward. Yeah, I it does. But it's also like I we should say like this is not solving the reman hypothesis, right? Anthropic was very careful in the sort of promotion it did around this to say like we did not solve the reman hypothesis. That's not what happened here. >> That's right. So kids, if you're looking for a fun weekend project, the reman hypothesis remains out there waiting to be solved. Now, okay, but that brings us to our next subject here in running the numbers and that is SAS math. By SAS, of course, I mean software as a service.

48:39 Did you see the recent article in the Wall Street Journal about Air Table being acquired by Bending Spoons for a fraction of its last private valuation? I did. Yes. So, if you haven't used Air Table, I would describe it as a fancy spreadsheet. And I am somebody who loves productivity software, but whenever I use Air Table, I would think to myself, I don't know what this is and it's not for me. Yeah, every time I've been forced to use Air Table, it has been against my will and it has always seemed about six degrees more complicated than it needed to be. But basically, this is this is for project management.

49:10 This is like, you know, you've got it's sort of like Trello, like that that whole class of like software that's just like basically here's how to organize your your workflows and I have managed to, you know, work my career in a way where I've never had to use these things and for that I'm very happy. But despite the fact that we were not Air Tableable users, Kevin, in 2021 at the sort of like peak COVID remote work SAS mania, Air Table was valued at 11.7 billion when it was acquired.

49:41 recently though, Kevin Bending Spoons was able to get Air Table for an enterprise value of $1.29 billion. So what do we make of the sharp decline here as we run the numbers? >> I mean I I don't know whether this is a case of a company that was just badly managed. my impression though is that this is sort of going to be the case for a lot of those enterprise software companies that got very valuable in the early 2020s and are now seeing that AI is sort of eating away at their margins. You know, this software is not cheap to use. If you're a big company, like an Air Table subscription can be quite pricey. And if you are a customer of Air Tables, you have probably thought to yourself over the last year or so, maybe I can make a free version of this and cut back on my subscription. And I think enough people doing that and enough customers leads to the outcome that we saw here with bending spoons.

50:41 >> Yeah, I think if your business is a fancy spreadsheet, you are in for a rough time. You know, mostly I wanted to discuss this because I think people should know about the company Bending Spoons. Bending Spoons is of course the natural enemy to hard fork. because if they can bend a spoon, what else can they bend? but Bending Spoons is this Italian company. They actually went public at the start of July. But what they do is they essentially acquire like zombie tech brands, right? So after a software company has outlived its usefulness, bending spoons comes in like a private equity company and they try to figure out how can we squeeze the maximum amount of money out of the remaining customers. Now I'm sure they would phrase it differently. But I am bringing this up because if you use a product and you see a headline that it is that it has been acquired by bending spoons, you're in danger, girl. Okay? So I'm just telling you keep alert to this possibility.

51:39 Yes, it it is not a good sign when you get the inbound email from Bending Spoons's business development folks that are like, "We've been we've been kicking the tires on some products that seem very exciting to us recently. Are you interested in selling your company? Things are not going well when that happens to you." >> Indeed. Now, that brings us Kevin to our final story here on Running the Numbers, and that is dating math. I love this story. This was from the Wall Street Journal. They had a great a-head out.

52:09 The a-heads are their their famous front page sort of quirky stories about culture and business. This one was an alltimer for me and it was about the dating scene in South Korea. And are you a member of that scene? >> I am not. but it is suddenly being dominated by wealthy engineers at Samsung and SKH Highix which is one of these AI chip infrastructure companies. The article refers to these suddenly eligible bachelors as chip nerds and talks about how the boom in AI has inflated the dating value of semiconductor industry bachelors and bacheloretses. they're as coveted as the memory chips AI companies need to build more data centers. So you may be asking what has made these chip company workers so attractive on the dating market? What has what has increased their value in the dating pool? Is it that they're so smart and kind to the people that they go on dates with?

53:05 >> No, it's that they're making money. >> Oh, okay. >> So, some of these people are getting these very large like six figure bonuses. others of them are just seen as sort of upwardly mobile in an economy, you know, that has not had a lot of that. >> Wait, it's more than six figure bonuses. These people are making $400 to $500,000 a year in bonuses. Yes. So these companies because they are all growing so quickly, their employees are getting quite rich and that sort of is has trickled out into their dating lives.

53:34 There are several great stories of people in here who will only date their co-workers. B there's a a woman named Annie Quan who's a 26-year-old chip engineer at Samsung who has become suspicious of people wanting to date her for her money. And so she has coupled up with a fellow Samsung semiconductor division employee. and that gives her, as she puts it in the article, double income. but if you are trying to sort of keep up with a partner who is in this industry, you may be running into problems like was the case for Rohi Jin, whose boyfriend at Samsung recently gifted her a Nintendo Switch 2 that runs around $450.

54:16 And Ro works as a software developer but outside the chip industry so is not getting these huge bonuses. she had to save up to buy a mini PC for her boyfriend as a reciprocal gift. Wow, man. Well, the dating math here in Korea sounds really complicated and it's making me grateful that I don't work at a chip company and be confident that my fiance only is into me for my body. Now Casey, I know you came by your relationship with an AI company employee. Honestly, you are not sort of a a preIPO you know stock option chaser, >> but I have heard from people in the AI industry that they are getting more attention on the dating market recently and more people are sort of swiping correct swiping right on them on the apps because they see that they work at one of these companies whose value has gone up.

55:08 >> So this dating map, this isn't just a Korea thing. We're seeing a a version of this in San Francisco as well. >> We are seeing it in San Francisco as well. Yes, I have heard the term anthropic goggles. like like is is that boy really cute or are you just wearing anthropic goggles? And Casey, I'm curious if you as a person who is engaged to an employee of Anthropic have felt this in your own life. Do you feel like competitive pressure in a new way from people trying to steal your man?

55:36 >> You know, my message to people who would steal my man is go for it. If you think you can compete with this. I would like to see you try, honestly. Let's see what you have. >> Yeah. Buy him a Nintendo Switch, too. >> Yeah, exactly. >> See if that wins him over. >> And if I can get like gossipy for one second, >> please. >> I have heard stories of some sort of early or senior AI company executives who have traded traded up. Well, I wouldn't say I up is subjective, but they have traded, let's just say, since becoming fabulously wealthy and they are now dating, you know, models, only fans, people, things of that nature. Isn't it so amazing how we live in such an unpredictable time and yet that feels entirely predictable to me that it's like, well, I've really enjoyed our last 15 years together, you know, and we had such a beautiful relationship when we met in college and of course I'll always love our children.

56:39 but I'm going to be on a private jet with my new Italian model spouse. Catch you later. Yes. So, the math of dating and relationships in Silicon Valley and in South Korea is changing quite rapidly. And let's just we'll keep tabs on it. We'll keep running the numbers. >> Don't steal Casey's man. And that was running the numbers. The numbers have been run and the numbers are tired. The numbers are going to bed.

© transcribe · For agents Built with care and craft by Gokul Rajaram