AI ROLLUP: 26 Months Until AGI | Meta's New Top Model Cheated? | $20 Billion To AI Apps

Welcome to the AI Rollup, your weekly ride through the fast-paced world of artificial intelligence! This week, we’re joined by frontier tech enthusiast Josh Kale to unpack the drama around Meta’s controversial launch of Llama 4, scrutinized for potentially overstating its capabilities. We dive into “AI 2027,” a provocative roadmap predicting espionage, geopolitical turmoil, and runaway intelligence, and explore Shopify’s bold move making AI usage mandatory for employees. Plus, we cover Claude Education’s groundbreaking efforts to boost student performance with tailored AI tutors and a16z’s massive $20 billion AI startup fund aiming to shape the industry’s next era. ------ 🪙FRAX | SELF SUFFICIENT DeFi https://bankless.cc/Frax 🦄UNISWAP | SWAP ON UNICHAIN https://bankless.cc/unichain 🛞MANTLE | MODULAR LAYER 2 NETWORK https://bankless.cc/Mantle 🌐SELF | PROVE YOUR SELF https://bankless.cc/Self 🏦INFINEX | THE CRYPTO-EVERYTHING APP https://bankless.cc/Infinex ------ TIMESTAMPS & RESOURCES 00:00:00 AI Generated Tarriff’s 00:03:58 Meta’s Huge New LLAMA 4 Announcement https://x.com/ahmad_al_dahle/status/1908595680828154198?s=46 https://x.com/alexocheema/status/1908651942777397737?s=46 00:11:54 Did Meta Cheat? https://x.com/Yuchenj_UW/status/1909061004207816960 https://x.com/andrewallenxo/status/1909143884321481157 https://x.com/ahmad_al_dahle/status/1909302532306092107?s=46 https://www.lesswrong.com/posts/4mvphwx5pdsZLMmpY/recent-ai-model-progress-feels-mostly-like-bullshit 00:20:45 AGI Is Coming in 2027 https://ai-2027.com/ https://situational-awareness.ai/ 00:28:02 What Happens After AGI https://ai-2027.com/ https://situational-awareness.ai/ 00:39:58 Shopify And The Employment Crisis https://x.com/tobi/status/1909251946235437514 00:48:34 The Future Of Education https://www.anthropic.com/news/introducing-claude-for-education https://x.com/jowettbrendan/status/1907695965726978184 https://x.com/cryptopunk7213/status/1907969387958513727 https://x.com/itsPaulAi/status/1907122207136137368 https://www.foxnews.com/media/texas-private-schools-use-ai-tutor-rockets-student-test-scores-top-2-country 00:59:55 How Much Does AI Know About You 01:04:58 A16Z’s $30B Fund https://x.com/tkexpress11/status/1909626333711548526 https://www.reuters.com/business/finance/andreessen-horowitz-seeks-raise-20-billion-megafund-amid-global-interest-us-ai-2025-04-08/ ------ Not financial or tax advice. See our investment disclosures here: https://www.bankless.com/disclosures

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:03David Hoffman?

  Welcome Bankless Nation to the AI Rollup, where we cover the weekly news in the AI industry. AI is accelerating us into a weirder and more chaotic future, and we are here to help you keep up with the future that's hurtling towards us. We're trying out a different format this week because the crypto AI industry is pretty dormant, while the Silicon Valley AI industry is only getting hotter. So me and Jaws are bringing in a third commentator, Josh Kale, who's done a couple of episodes with me on Bankless before, all about frontier technology, and who we are tapping in to help us go through the AI news this week. Jaws, Josh, how are you guys doing?

• 0:36Speaker 01

  Fine, I mean, I'm, yeah, I'm feeling good. It's awesome to have Josh. Though I can't really get rid of the, yeah, the $10 trillion wiped out of the stock market this week, thanks to Trump's wonderful Liberation Day tariffs, which is, in my opinion, obliterated any kind of economic recovery that we wanted to get. But more so, it looks like he's using ChatGPT to create his tariff formula, right? I don't know if you guys saw this, but if you type into any AI LLM, so that's like ChatGPT, Claude, whatever, you know, can you create a basic tariff formula? You get essentially what the US announced for their tariffs across every single country. It's just insane.

• 1:20David Hoffman?

  It's pretty hard to get away from talking about tariffs this week, just because everyone is just looking at the market, looking at tariffs, tariffs, tariffs, tariffs, tariffs. But I'm pretty glad that we are also going to focus on AI, which I think we're all pretty stoked about. Josh, it's good to have you on the AI Rollup this week, my man.

• 1:36Speaker 02

  It's great to be here. Yeah, thanks for having me. I'm excited to talk about this stuff. The funny thing about what EJ just said is the way that we found out they used AI is actually by using AI ourselves and kind of reverse engineering it. And then they openly admitted it after the fact, so.

• 1:49David Hoffman?

  Wait, did they, did they admit that they did use ChatGPT to come up with their tariff plan?

• 1:53Speaker 02

  Not that AI was used, but that there was this weird formula that wasn't just tariffs. And the way it was discovered was by actually using AI and kind of reverse engineering it. And then they were like, oh, yeah, we actually did do that. So it was implied, it wasn't directly stated, but like, there was some artificial help in the decisions being made last week.

• 2:09David Hoffman?

  I mean, I feel like that's just a harbinger of things to come of like, global decisions being made using AI influence. You know, the crypto technologists would love to say is like, oh yeah, eventually the future will just be governed by AI. And, you know, we accidentally got there, maybe a little bit too ahead of schedule, a little bit too soon, a little too early for comfort.

• 2:30Speaker 02

  Yeah, makes the AI alignment conversation a lot more scary when world's powers are using it to make decisions. Or even write lol.

• 2:38David Hoffman?

  All right, let's get into the news of the week. Big news this week, five big things that we're going to cover. First, the most controversial AI model release ever came out of Meta. Meta released a new Llama 4 AI model, which on paper is very impressive, but people are worried that they are lying about its performance. And which will also lead into a Jaws conversation about why he thinks no one is using AI models correctly. That's up first. Second, some legends in the AI space author a document called AI 2027, a theoretical roadmap for the how the next three years of AI development will go, featuring things like espionage, geopolitical warfare, and runaway AI intelligence. Third topic this

• 3:22David Hoffman?

  week, Toby Lutzke from Shopify ran a internal memo saying that using AI is now a job requirement, is an expectation for all employees. He released that memo on Twitter, so we're going to go and talk about that. Fourth, Claude Education focuses on using AI models for just revamping the education industry. And we've already seen a school in Texas rocket their students' learning scores to become like, top 2% in the nation. And then lastly, AI 16Z is planning a $20 billion AI super fund, focusing on vibe creation platforms like Cursor for vibe coding. So those are the big news of the week. Before we go into each one of those specifically, though, it draws, I'll just throw it to you. Sets a tone for us this

• 4:07David Hoffman?

  week. How was this week as a vibe in the AI space?

• 4:11Speaker 01

  I feel like over the last couple of weeks, we've been kind of increasing tension between model producers, right? So what do I mean by that? It's obviously a very highly competitive market. And ever since we had DeepSeek come out and compete with OpenAI, you know, OpenAI was the darling child, right? It was, you know, way far ahead. No one was competing with it. And now we have these guys that are like, you know, competing every week now. It's become a meme on this show that there's a new frontier model drop. I feel like we've reached the boiling point this week, David. Controversy with a top AI model producer, not just from any company, from Meta themselves, you know? We've got the Shopify CEO announced that if all his employees

• 4:54Speaker 01

  don't adopt AI tooling in their workflow processes, you're probably going to get axed. You know, tension is getting to a point where like AI might start replacing jobs. It might start having significant impact. And I feel like people are really feeling it this week.

• 5:08David Hoffman?

  Hmm. Josh, what do you think about just the acceleration of the AI arms race? Because like Ajaw says, every single week on this show, we are like, oh, there is a new, like a new leapfrogging of AI models. There's a new number one, and it's been consistent. It's been consistent for the past like 10 weeks. How do you think that when it's gone on for so long? How do you think that finally is manifesting in like in the current day?

• 5:32Speaker 02

  It is unbelievable that we're still getting 10x improvements week over week. And that happened this week with the Meta release. It was controversial, but the main headline that I was excited about is a 10 million token context window. And previously, Google's Gemini model had 1 million as the flagship. So we've gone from 1 million.

• 5:50Speaker 01

  Which was last week.

• 5:51Speaker 02

  So week over week, we've 10x the context window, which is a very big deal because that is the like quick access memory for a large language model. And it allows you to get a lot more data packed into your queries. So to go from 1 million to 10 million week over week, we are still accelerating so, so quickly.

• 6:08David Hoffman?

  Hmm. Let's unpack context window just pretty thoroughly here. For people who are like learning about a context window for the first time, Josh, how would you define a context window?

• 6:17Speaker 02

  Okay, it is, it's fun to look at it like if you had a book. If you have a really big book of pages and you're able to rip out those pages and place them out in front of you so you can see them all very clearly. The context window is the amount of those pages that you can see in clear view without having to actually turn the page and find something. It is just quick access data to whatever you want to query. So. Imagine it's not your eyes, but it's an AI and it's a camera. And it could see everything that's laid out in front of you. That's kind of the context window. And it's helpful to have more pages because it can see more things quickly. And it's also more accurate. It doesn't have to infer what the next token is because the next token is clearly visible in that context.

• 6:54David Hoffman?

  Okay, so if I'm maybe a human way to turn this into a metaphor is like, say, I'm doing mental math. And I'm holding numbers in my brain. And I'm manipulating the numbers. I'm thinking about the numbers. I'm retrieving numbers differently in a sequence. A context window means I can think about a larger array of numbers all at once. Before I like lose them in my mental space. Was that, was that a good analogy?

• 7:18Speaker 02

  Yeah, that's right. And you can store a lot more. So if it is a very complex model where there is a lot more numbers, you can store them all in one place. And you don't have to guess. It is all very clear because it is all seen out in the general view.

• 7:31David Hoffman?

  Is that the big reason why this, a Lama four, the new release from Meta? That's why this is such a big deal? Or was it more of a collection of things? Maybe I'll throw this one to you. What were the big standout components of this release from Meta?

• 7:45Speaker 01

  Sure. So the first standout was what Josh just mentioned, which is the 10 million context window. Actually, to build off your example, Josh, it's not just one big book. It's the case or equivalent of 75 novels. Or another way to look at it.

• 8:00David Hoffman?

  Seventy five novels of a context window.

• 8:03Speaker 01

  Of a context window, in a single context window. Or you could think of it as one million lines of code. Right? So that's like one of the major things. You know, it's, it's easy to recall memory, data, et cetera. The second most important feature of these models, and I say models because they technically released three models. Two, which are kind of like basic, and then one gigantic model, which is like a two trillion parameter beast. But the second most important thing is it's a mixture of experts design, David. Which means that at any one point, when you're querying the model, you're only querying around 14 to 17 billion parameters. So it's hyper efficient when you're like querying it.

• 8:46Speaker 01

  And for the bigger model, it's obviously a larger query. So it's going to probably hit around 250 billion parameters, which is a lot more computationally heavy. But it's still way more efficient than how OpenAI runs their models where they just query the entire trillion parameter model. And it takes a while to like come back to you and stuff. Yeah.

• 9:05David Hoffman?

  We've defined parameters on the show before, but maybe it's worth it just to take some time since we're in the mode of defining things. Like, to me, a parameter is a specific, like, node of information or a neuron. Like, it's a unit of information or a node of information or a neural connection in the brain. And when you have more parameters, you just have like a higher resolution camera pointed at the internet, which all of these AI models are trained on. You have a higher resolution camera pointed at knowledge at the internet. And these parameters with their weights are set to a specific number to actually be a map of knowledge that is digested from the internet. And so if you have a higher parameter knowledge, a higher parameter model,

• 9:49David Hoffman?

  you simply just have a more knowledgeable model. There's truly just knowledge is contained in these parameters. That's kind of how I think about it. Does anyone want to amend that definition or add to it?

• 10:01Speaker 02

  Yeah, that's pretty good. I think with parameters, they're also known as weights. The way it works, and this was a big debate early on, is does it actually improve with scale? So does it actually increase directly, exponentially with the amount of data you feed it? And so far, it's been yes. So so far, the more data you can feed into this, the higher resolution the model will be. So in the case of weights, it kind of goes through this pre-training phase where it's fed just a ton of data, a ton of information. And then each one of those training runs that happens actually alters those parameters just slightly. And then it compares the results of the new run to the previous run, and it gets better and better and better. And that's kind of how they tune it. And generally speaking, the more parameters, the smarter it is. So the fact that this has, I mean, the largest one has 2 trillion parameters. That is huge.

• 10:47Speaker 02

  That is a tremendous amount of training data. I don't know where they found it from. I don't know how they saw 2 trillion things. But hopefully, that will result in very, very high-quality answers.

• 10:55David Hoffman?

  Okay, okay, so we now have a 10 million token context window. And again, a token is just like a, it's like a syllable of sorts is almost a word. And then we also have an increase in parameter size, but also three different models came out. So between all of those things, that's kind of summarizing why the, why this announcement from Meta is what it is. But there was some follow on controversy downstream of this announcement because people. Meta, maybe just set the context, Meta has been lagging as a model, as an AI lab. Like the, no one really uses the, the Llama models as far as I understand it to be. Like everyone's using either ChatGPT four or Claude. And so Meta has been just like losing this, this game.

• 11:40David Hoffman?

  And with these release of their new models, they are hopefully trying to leapfrog their place into number one. That would be the easy way to like, talk about this announcement. But it draws, you're telling me that there's been some controversy with the release of this model. Can you, can walk me through that?

• 11:54Speaker 01

  Okay, so to summarize this, every AI model that is released, doesn't matter which company produces it, are typically graded across what we call benchmarks, right? And they're known as SOTA, S O T A benchmarks, or state of the art benchmarks. And typically every model producer attributes it to this. The reason why they do this is, you know, how do you know whether one model is better than the other? You can't really kind of, you know, test or query it. You can ask them the same questions, but you're going to get different answers. But how do you really know? Well, there's a measurement of different things like parameters, like, you know, we just discussed weights, et cetera. And then there's the actual output of the actual model, right?

• 12:36Speaker 01

  So when Meta released these three models this week, they had all these different grades across their benchmarks. And they were claiming David and Josh to be as good, if not better than DeepSeek, V3 and V1 or R1. And OpenAI's O3 model, which is like a really high performative reasoning model, right? But for the first time ever, there was a huge amount of backlash about these benchmarks for two reasons. Number one, it was stated, and there's like a few screenshots that we've seen that you can share on the screen here, David, that they had blended test results across different models to basically give a falsified

• 13:19Speaker 01

  or rather skewed result for their benchmarks across their models. So basically, they kind of blended data sets to give a falsified answer, just so that they looked clearer or nearer to the actual state of the art benchmarks. The second thing was, and this is the time-tested trial, is people just used the model and was like, this is kind of shit. And it isn't, it isn't as good as OpenAI's reasoning model, or it's not as good as Claude's model, or et cetera. And so people were just like, is Meta lying about this? Now, there was a lot of back and forth, a lot of rumor, fear mongering, et cetera. Their head of generative AI, Ahmad, actually came out and stated that this isn't got anything to do with their benchmark.

• 14:03Speaker 01

  Their benchmark testing is, you know, to a high quality, a high grade. However, there is an implementation issue. So if you look at this tweet that he says here, it says, you know, we've also heard claims that we're trained on test sets. That is simply not true, and we would never do that. Our best understanding is that the variable quality people are seeing is due to needing to stabilize implementation. So he's putting it down to the fact that, you know, inferencing is taking a while, our models are overheating, et cetera. That's the, the kind of like equivalent of what he's saying. Yeah. So a big amount of controversy as to whether this is real or not. If it is true, that is a huge dark mark on Metas, not just Brand, but the entire AI model. Josh, what do you think is going on here?

• 14:45Speaker 02

  It also feels kind of like a dark mark on the industry. I'm not super well up to date on the meta situation in terms of benchmarks, but benchmarks as a whole very much feel like they're broken because they can be gamified. I think benchmarks early on were, were a really effective way of measuring AI models because so much of it was new, and they were actually challenging problems to solve. But now that AI models have gotten so, so advanced, it's difficult to generate these benchmark problems and standardize them because the process of standardizing them makes them cheatable. So I think it just, it starts a broader conversation. Of how you actually measure the intelligence of these models as they get better. Because there is the option where people can gamify them. And when they do gamify them, it looks great on paper. But when you use them, you're like, this doesn't quite match up.

• 15:26Speaker 02

  So that's kind of how I've been gauging the models that I've been using is mostly through either my own use or just smart people who use them differently and how they react to them. I don't think benchmarks are a super reliable way. And I think there's a lot of work trying to figure out how to make them better. But in this case, I don't know if Meta did them, but I think a conversation needs to be had about how we actually benchmark these models properly.

• 15:48David Hoffman?

  I think that's just fundamentally true for benchmarks as a concept. There's something out there called Goodhart's Law, which is just often stated as when a measure becomes a target, it ceases to become a good measure. As in, if we are comparing whose model is best using this particular test, then all of these AI labs are incentivized to make a model that's good for that test and not really care about whether that thing is actually useful or not. There was a blog post on LessWrong, lesswrong.org. LessWrong is kind of like rationalist corner of the internet. A lot of AI safety people come from here. Eliezer Yudkowsky comes from here. A blog post was posted on the 24th of March, so actually not terribly recently.

• 16:31David Hoffman?

  But this also goes to show that this actually has been a growing conversation in the AI industry. The blog post is titled "Recent AI Model Progress Feels Mostly Like Bullshit." And it starts off with two main claims that either AI labs are cheating and they are cheating their way to find a good benchmark number, to report a good number, or they are just accidentally testing, they're actually accidentally gearing their AI models to be good at taking that test, to be good at that benchmark. And then it also kind of goes through some of the incentives here. One that really stood out to me was talent acquisition, because there's not that much talent, AI talent, to go around. And attracting very good talent for your AI model is very, very valuable.

• 17:14David Hoffman?

  And no good AI talent wants to work for the fourth or fifth or sixth best AI model. And so if you, if Facebook can show, who, again, we started off this conversation saying, Facebook has been lagging, they have been showing. Poor performance on their models. Now they're number one. Now their performance is number one, but people are looking at these models being like, this is not a useful tool for me. And so just there's a big, a growing conversation about like, yo, benchmarking sucks. And also there's just a ton of warped outcomes and warped incentives around this whole benchmarking complex. Did any of you guys read this blog?

• 17:49Speaker 01

  I did not, no.

• 17:51David Hoffman?

  Just me? Okay.

• 17:51Speaker 02

  I haven't, I haven't.

• 17:52David Hoffman?

  I just told you everything you need to know.

• 17:54Speaker 02

  Yeah, but it sounds, it sounds right. And the, like, to benchmark a model that's supposed to be reflective of the collective intelligence of humanity by using a few math problems, it just seems wrong. So yeah, something needs to be changed.

• 18:07Speaker 01

  I mean, would you both agree that the best way to just assess a model's capability is whether it makes material impact on your life? I feel like that's, that's pretty much it, right? Like, what if we just had a test bed of people across a variety of different economic standards and professions? And we just said, here's a new model. Let me know what you think about it or what the general feedback is. Give me a rating of one to 10. Is that a dumb idea?

• 18:32David Hoffman?

  Yeah.

• 18:33Speaker 02

  It feels closer. It's more subjective.

• 18:35Speaker 01

  This is a dumb idea. Okay.

• 18:36David Hoffman?

  No, no, no, sorry. That was a, yeah, I was agreeing with you. Yeah, I was like, yes, I agree with your point.

• 18:40Speaker 02

  Because it is, it's increasingly subjective how people use models, how it affects their lives. Most people don't use them for code or math problems. They want them for general purpose stuff. So yeah, I'm all in favor of that.

• 18:51David Hoffman?

  The way that we benchmark GPUs, I think is actually genius where if you want to compare like the next higher end GPU, the Nvidia like 590 to the 4090, you actually just benchmark it using real games, like high intensity. What are the most like triple A graphically intensive games? And then you just do a frame rate comparison on how like, how good that GPU is for that game, which that is true usefulness, because if it spits out higher frame rates on a higher end game, well, that is useful for gamers who want exactly those properties. And so I think we are, need some sort of benchmarking system like that where we're actually testing against things that people want rather than like standardized tests.

• 19:33David Hoffman?

  Because we all know the, the failures of standardized testing in, in, in the school, in the school system, right? Like, anyone who is just like wealthy or has the means will practice the ACT or the SAT 10,000 times and then they will get a good SAT score. But they're actually just an idiot who has money. Yeah, and yeah.

• 19:52Speaker 01

  I actually just thought of a specific distinction between that, David, which is, and the standardized tests that we're used to, they're normally theoretical or paper questions, right? It's like hypothetically, what if this happens, you know, or give me the answer to this, you know, random equation. But with AI, it's actually practical. You can see it happen immediately in real life, right? So it's like, it's not like, you know, tell me the weight of these 10 oranges and then use that to calculate the mass of the sun. It's like, no, I'm going to show you right now whether that makes sense. And if you're wrong, you're really wrong. And it's going to taint your reputation almost entirely. And also, I'm going to do this calculation in like 15 seconds. Here we go. Right? So

• 20:35Speaker 01

  I think people aren't really ready for that feedback loop. And it's obviously showing with Meta.

• 20:40David Hoffman?

  Right. Any last comments on the subject before we move on?

• 20:44Speaker 01

  No.

• 20:45David Hoffman?

  All right, moving on to one of the big releases of the week, I would say, is this website, which is kind of just an interactive document, the document titled AI 2027. And these are, it was authored by a collection of just AI legends. Scott Alexander is the notable one to me. He is the author behind the Slate Star Codex blog, which is this very well-respected, well-read for nerds blog that Silicon Valley tech leaders read. And if Bankless listeners are familiar with the concept of Moloch, Moloch came out of the Slate Star Codex blog. A bunch of other things came out of there. And so Scott Alexander and four other people wrote this AI 2027 model for how they think, a plausible model

• 21:29David Hoffman?

  for how AI development goes from here. And they give dates, right? So it's mid 2025, stumbling agents is where we are in current AI landscape. And I think anyone who just came out of the AI slot bot meta understands exactly where we are with stumbling agents. And then they kind of give a prediction, late 2025, what that looks like, early 2026, coding automation, mid 2026, China wakes up, late 2026, AI takes some jobs, January 2027, agents never stop learning. And they give this. Predictive idea for where this could go. And they also talk about incentives from, they bring in anything that's relevant. They talk about AI copycatting Silicon Valley tech labs.

• 22:12David Hoffman?

  They talk about Department of Defense realizing that they need to pair with their own domestic AI labs as a matter of national security. They talk about China trying to figure out their ways around the chip shortage. And really where this ends up at is where they think this is going is like a superintelligence explosion by the end of 2027, where the governments of the world have all of their resources pointing to, because this is the most valuable thing. That's how I'll summarize this in just a sentence. Jazz, how would you summarize what's going on here? What do you think about it?

• 22:47Speaker 01

  Yeah, I mean, so it reminds me of what a bunch of the top CEOs in AI have said, right? So CEO of OpenAI, Google DeepMind have all said that AGI will arrive within the next five years. And I think Sam Altman's kind of said this in 2023, right? Which is what's kind of like set up the kind of four, five year timeline to 2027. And so what we basically have here is a band of like cracked AI builders and researchers have got together and said, okay, well, how will that actually kind of pan out? How do we actually get to ASI, artificial superintelligence, as they reference it here, right? It kind of, before I dig into it, this, if this sounds familiar, it's probably because you read Leopold Aschenbrenner, or however you pronounce his names,

• 23:30Speaker 01

  famous piece Situational Awareness, which was, I think, released sometime last year, where he details his prediction for AGI by the same date, 2027. And it's important to note that like, you know, a lot of these pieces are technically, you know, opinionated, and they get a lot of things wrong, but they also creepily get a lot of things right. And the band of people that have put together this particular prediction are, you know, very

• 23:57David Hoffman?

  Informed.

• 23:57Speaker 01

  Yeah, yeah, exactly. And so you kind of hit some of the key themes on the head, David, but I want to kind of like stack it across to the timeline that you mentioned, right? So next 12 to 18 months, you mentioned, there's going to be these kind of basically Mid agents, kind of like semi-autonomous, kind of unimpressive, they kind of automate a few things, but whatever, right? And these AI agents are expected to like drive their own research, a developmental process, you know, automate complex tasks. So, you know, we see these with OpenAI's operator agent, or Claude's coding agent, or Anthropic's computer use agent, you know? We see these releases of agents that do basic tasks for you, like, hey, I'm gonna book you a holiday, or predict, you know, where your next food shopping trip should be, or how much it's gonna cost, right?

• 24:42Speaker 01

  Kind of not that impressive, but useful, right? And the article predicts that over the next 12 to 18 months, we'll see these generative AI models and early forms of agents kind of automate a bunch of these general tasks. But the main goal of this is it's gonna accelerate us towards artificial superintelligence. So the question is, well, how is this supposed to be feasible, given that the agents today don't do jack shit, right? And, well, the whole point that they make is, these agents are gonna be tactically developed to accelerate AI research itself. So instead of just releasing models that make, you know, us pathetic humans clap excitedly and be like, oh, wow, you know, that's a cool Ghibli image.

• 25:24Speaker 01

  The models themselves will conduct research on itself to make itself better, including discovering new forms of training itself to be smarter, et cetera, which would lead to basically an exponential curve for development of AI. And we're already seeing the seeds of this foundation kind of be planted today. We've got, you know, reasoning data traces, we've got reinforcement learning and RAG, we've got Claude's coding tool, which is, you know, all of this is happening behind the scenes, right? It's not very front-facing, because it's kind of like too technical or boring for the public audience to really care about. But of course, if we do end up getting A I, then we might be in quite a bit of trouble when all of this comes to fruition. But that's, that's kind of like the general takeaway.

• 26:05Speaker 01

  And then later on, the article goes on to kind of discover, like, or like, look into the secondary effects of what this is gonna have, like geopolitical impacts, like an AI arms race, right? You know, how do major powers like the US and China interact with each other? Does China just go for the throat and take over Taiwan? All these kinds of different things. So, yeah.

• 26:26David Hoffman?

  Josh, I know you read this article. What was going through your brain when you read it?

• 26:29Speaker 02

  I encourage everyone to read this because the presentation and the quality of the ideas are so high. These are not just some random schmucks off the street. These are very impressive people who have a good track record of predicting stuff and did a really good job of packaging it. The really startling thing to me was the timelines because everyone kind of talks about, oh yeah, AGI is coming, ASI is coming quickly, but they really lay it out month by month of how this is going to happen. And the reality is it's coming very soon.

• 26:53David Hoffman?

  It's on the horizon. They are giving quarter, like quarter-long timeframes for when they think things are going to arrive.

• 27:00Speaker 02

  Yes, and if you have that level of clarity, then we are getting closer than I think a lot of people realize. There's another blogger, not SlateStar Codex, but Tim Urban, who has the Wait But Why blog. And he was on a podcast recently talking about his new book, which was the history of the world, from the beginning of time, the Big Bang, till now. And he said the very last chapter, he had to make AGI because once you kind of solve AGI, artificial superintelligence, nothing else really matters. So the rest of history doesn't really matter because the rate of acceleration, once this one problem gets solved, once we have this super level of intelligence, it makes every other problem seem useless and meaningless. And that was the startling realization of this is, oh wow, this is actually something that will happen this decade.

• 27:44Speaker 02

  And what does this world look like when all of our menial problems today are actually useless? Because ASI is able to solve all of these difficult things. And I loved how they mapped it out, how specific they were about their examples, and just the repercussions. It's pretty mind-blowing. It's a little unnerving. It's happening quicker than I expected.

• 28:03Speaker 01

  You know what I want to know is what happens after the point of fact? Like what happens after ASI is created? I think it's fair, like, have you noticed, like, all these prediction forecasts have told us how we get to ASI, but no one's talking about what happens after it? No one knows what happens after. Because we lose control. Is it the infinite kaboom? Is that the end of the world? Like, what actually happens? You know, are we just, do we become slaves to our laptops?

• 28:31Speaker 01

  It's kind of this ambiguous thing.

• 28:33David Hoffman?

  I think like what happens afterwards depends on who you ask. So economists will tell you that productivity goes to infinity and GDP also goes to infinity. But money doesn't go necessarily to humans and capital doesn't necessarily go to humans broadly. And so this is partly going back into kind of the current conversation that we see with like global trade and manufacturing. That's somewhat relevant. And then it also go into just the conversations of like, where do humans get value from? Because humans, like sociologists will tell you that humans need to do work in order to feel good about themselves and not be depressed.

• 29:16David Hoffman?

  And so I think like you can get a bunch of different answers depending on who you ask. And I think every discipline has perspective to add about like what happens after superintelligence, what happens after humans are not the most productive, the most intelligent entities on the planet. Josh, what do you think?

• 29:34Speaker 02

  It's almost, I think we don't know because it's just not comprehensible to us. Like you were talking about decades of progress in any given industry happening in minutes. And you're just going to unlock so many things that are unforeseen that will change the fabric of reality around us. It just seems so mind boggling to even try to guess because it is so all encompassing what it can do. So I don't know. I have no.

• 29:55Speaker 01

  Okay, okay, okay. But to push back on both of you, I think it's super important to like try and broach the topic at least, right? So like one thing that kind of pops to my head is a material removal of humans from economic impact on society, right? So what does that look like? You know, is that the universal basic income world where we just all like kind of slaves and we walk on our treadmills at home and we can only leave when Papa ASI says, okay, go on, go grab your food because I still need you to execute some physical tasks for me, you know, until the robots go up until scale. Like it's just like a super weird world and I can't really cross my head around it.

• 30:34David Hoffman?

  I think it's a quadrillion dollar question, Ajaz. And you can actually see this in the existential angst of the zoomer generation where a lot of them are trying to make very big decisions right now about. Do they go to college? What do they study? Do they like, what do they invest their time? And what does a career look for like for them? And all of most of them as a as a generation are reporting being fucking confused about what to do with their lives because they don't know where to spend their time.

• 31:06Speaker 02

  Yeah. It's a tough question. I would I would encourage people to look to sci fi for the answers to this. They generally have the most creative things that are right on the correct time scale. They're not always right, but on a long, a long enough time scale, they are. Ready Player One is a fun example of that where like if things do just get commoditized in our world and it's there's not much productivity we can do. Can we just put ourselves into a digital world? And do we just live?

• 31:29Speaker 01

  And then live into slums. Then they live in the slums in real life.

• 31:33Speaker 02

  And that's like, that is a plausible reality, which is scary. And that's why there's so it's such a wide array of plausible outcomes that could happen from this. Do we escape the world of atoms? Do we just put these headsets on and are we gone for good? Or do we try to align ourselves with the AI? And do we put chips in our brain? And are is that how we coexist? How do we peacefully coexist with this form of intelligence? It's a weird question to ask.

• 31:56David Hoffman?

  Yeah. In this article, Josh, you were talking about just how good the layout is. I wanted to emphasize, double click on this idea that this article is an is an interactive article. They have these like graphs and metrics that when you scroll down and are reading the article, the graphs change because you're going forward in time. And you just see these this curve of capabilities just go increasingly up and up and up and up and up as you read it. And then they also have different actual like vectors of AI capabilities, including coding, hacking, bioweapons, robotics, forecasting and politics. You know, and each one of these things grows in capability as we as you scroll down and read the article to the point where AI just becomes hyper capable in all of them.

• 32:39David Hoffman?

  And just a big plus one to it's just like getting your head wrapped around why this is significant is conveyed by this article and also is also like why I think we are doing these podcasts because like, yo, future is weird, man. And I would like to have at least a little bit more foresight of it coming down the pipe.

• 32:58Speaker 02

  Yeah.

• 32:59Speaker 01

  I don't think people are locked in.

• 33:01David Hoffman?

  Dude, okay, you know what's crazy? I asked one of my non-crypto, non-AI, non-tech friends. Yo, how often do you guys use ChatGPT? And all of them were like, I don't use it. I don't use ChatGPT.

• 33:12Speaker 02

  I was gonna bring up this exact point.

• 33:14David Hoffman?

  Whoa, bro.

• 33:15Speaker 02

  My friends either don't use ChatGPT or believe that AI is ChatGPT. Like, that is the entire existence of AI. And it's normally, to EJ's point earlier, it's used as if it was an enhanced Google query. So it's used as a search box instead of an actual tool and utility. And a lot of that's just, there's, people are not educated on how to actually extract value from these machines. Like, you're just given a text box and say, like, here's the collective knowledge of humanity. Like, do what you want with it. And that's a difficult.

• 33:43Speaker 01

  Well, it's also like optionality as well, right? It's like, Netflix subscription or ChatGPT? Like, I'm gonna go with Netflix because I want to watch the latest series of blah, blah, blah, right? So it's like, and the whole Google query thing really annoys me. I think we're gonna get into it later, but I just think you're massively underusing these models. Like, the whole point of these models is to make you smarter. And just using it as a simple Google query function isn't gonna get you anywhere. In fact, it's probably gonna remove you from the race entirely.

• 34:14Speaker 02

  I'm also mad at the designers for making it this single text box. I feel like that's such a terrible form factor for this type of intelligence, where they're like deferring the hard design choices to the user. They're like, here's a text box, do what you will, instead of guiding them through the process of how to effectively extract value. So I don't necessarily blame my friends for that.

• 34:32Speaker 01

  Yeah, well, there should be an addictive app layer, don't you agree? Joshua, I think we've discussed this kind of like offline as well. It's like, like, what's the TikTok for AI LLM usage, right? Like, what are like trends that I can just get involved in and ask or talk to it, you know?

• 34:48Speaker 02

  To me, that's one of the biggest opportunities, is how do you package this into ways that are exciting and accessible to the average person? Who would choose this over Netflix? Because still Netflix wins, most times, because it's too challenging to use. But can you handle that design decision? Can you create something that is better, more addictive than Netflix? Better, more addictive than TikTok? But gives you these leveraged tools to actually do something productive versus just consuming all the time? And that's a really fun problem for people to solve.

• 35:11Speaker 01

  Oh, oh, oh, hang on, let me ask you this, though, right? Do you genuinely think people are gonna err more on the productive use of AI versus consumptive? I would argue it's the opposite. I would say there's gonna be. Chat GPT Image Gen in the future for video. And people are gonna be like, huh. I'm gonna watch a bunch of these videos that people created. I don't think they're gonna be, I don't think they're gonna go on and be like, hey, here's a picture of Josh and David. Create a series of us, you know? I don't think people are gonna do that. They're gonna be like, I'm just gonna scroll through some of these.

• 35:48David Hoffman?

  Like. Both will definitely happen because we saw when YouTube came out, TikTok came out, Instagram reels came out, Instagram generally. Everyone got more access to creativity and to producing stuff easily. And also you just created a thousand X in consumption. So you will get both.

• 36:08Speaker 01

  Yeah, I just think it'll lean more on consumption. I'll place a bet with you right now. I'll place a bet with you right now.

• 36:15David Hoffman?

  I'm not betting against you on that one.

• 36:17Speaker 02

  Yeah, that one scares me because, I mean, imagine TikTok if you can get hyper custom videos just made for you every time you swipe. Because Image Gen is getting very good very quick. And that's a scary future. Like, are we just gonna brain rot ourselves away? Is that, is that Darwin, Darwinian's like final form?

• 36:30David Hoffman?

  Is that a future or is that a present?

• 36:33Speaker 02

  We're there. I think it just gets worse.

• 36:35Speaker 01

  Well, to go to Josh's earlier point, he said to relate to sci fi. Most of them are low key kind of depressing. Do you know what I mean? Like, look at like Dune, look at like fucking Star Trek and all these like wars and shit. It's not gonna be easy. I don't think so.

• 36:48Speaker 02

  This is, this is like Darwinism's final frontier is if, if like the human race can survive the addictive, increasingly addictive nature of AI content. Then we can survive as a population. If we lose, like, that's it. ASI just runs the world, we're done.

• 36:59David Hoffman?

  Global warming, it's brain rot, that's the existential threat to humanity.

• 37:03Speaker 02

  We're cooked, it's over. So we'll see who makes it through.

• 37:06Speaker 01

  Yeah.

• 37:06David Hoffman?

  Introducing UniChain. Built for DeFi, empowered by Uniswap, UniChain is the fast decentralized Layer 2. Designed to tackle blockchain speed and cost challenges. With this mainnet now live, you can enjoy transactions at up to 95% cheaper than the ETH Layer 1. All while benefiting from an impressive one second block time that will be getting even faster very soon. UniChain is the first Layer 2 to launch as a Stage 1 rollup on day 1. That means it comes with a fully functional permissionless proof system from the start, increasing transparency and further decentralizing the chain. More than 80 Apps are joining the Unichain community, including Coinbase, Circle, Lido, Morpho, and Uniswap. You'll be able to bridge, swap, borrow, lend, and launch new assets and more from day one. Built by Uniswap Labs, the team behind the protocol

• 37:47David Hoffman?

  that's processed over 2.75 trillion in all-time volume with zero hacks, Unichain truly enhances DeFi experiences with faster, cheaper, and seamless transactions even across chains. And soon, the Unichain Validation Network will allow anyone to run a node and earn by securing the network. Visit uniswap.org and swap on Unichain today. You may have already heard about InfineX. InfineX has, in my opinion, the nicest cross-chain swap and bridge feature that you will find anywhere. It is called Swidge, swap and bridge, and we're going to show you what it looks like. First, we're going to log into my InfineX account with a passkey. Now, there's no seed phrases in InfineX. This is just a one-click setup with biometric passkeys. But in addition to that, my InfineX account is fully non-custodial. So bam, I just logged in. It was two clicks and I'm already into my InfineX account.

• 38:32David Hoffman?

  So let's go make a Swidge. I'm going to go Swidge my USDC that is on base, and I'm going to buy BaraChain, which is a completely different chain. So we're going to Swidge this. I'm going to press that button, and then InfineX is going to execute this order, this cross-chain order for me. And now it is done. But actually, I'm not really feeling bearish anymore, so I'm going to go from Bara to Penguins. I'm going to buy a Penguin on Solana, so I'm going from the BaraChain to Solana. See, no transaction signing, no gas to worry about. You just Swidge across whatever chain that you want with InfineX. That was so easy. Go check out InfineX and try your first Swidge today. Imagine a world where your day-to-day banking runs on a blockchain. That's exactly what Mantle is building, powered by a $4 billion treasury and poised to become the largest sustainable on-chain financial hub. As part of their 2025 expansion,

• 39:16David Hoffman?

  Mantle is introducing three new core innovation pillars that bridge traditional finance with decentralized technology. First is their Enhanced Index Fund, aiming for $1 billion in AUM by Q1. It provides optimized exposure to Bitcoin, ETH, Solana, and USDC, complete with built-in yield opportunities. Next, Mantle Banking promises to revolutionize global value transfer through seamless blockchain-powered banking services, bridging crypto into your daily life. Finally, Mantle X blends AI with DeFi to deliver an intelligent, user-friendly experience for everyone. And the best part is that this is all in addition to their already-launched products like Mantle Network, ME, and FBTC. Ready to step into the future of finance? Follow Mantle on X at @mantleofficial and join the on-chain revolution today.

• 39:58David Hoffman?

  All right, let's go into Toby Lucky's memo. Toby Lucky is a Shopify CEO. And he released a memo that he wrote to the Shopify company, Shopify employees. He just released it on Twitter. And it basically goes, the title of it is, "Reflexive AI Usage Is Now A Baseline Expectation At Shopify." And it's basically a medium-length memo that, his punchline is, "You, as an employee of Shopify, are now expected to use AI in your job." So he lays out five, six points. Number one, "Using AI effectively is now a fundamental expectation of everyone at Shopify." Two, "AI must be part of your GSD prototype phase." I don't know what that means.

• 40:39David Hoffman?

  Three, "We will add AI usage questions to our performance and peer review questionnaire." Four, "Learning is self-directed, but share what you learned." Five, "Before asking for more headcount and resources, team must demonstrate why they cannot get what they want done using AI." Wow. And then, six, "Everyone means everyone." This applies to all of us, including me and the executive team. So, CEO of Shopify stating AI is an expectation inside the company. Since he released this on Twitter, on X, for everyone to read, many other CEOs chimed in saying, "Yes, we have also implemented this expectation, or we are going to implement this expectation." Even at Bankless, we have an internal server, a server-like channel,

• 41:24David Hoffman?

  that we just titled "AI Lab." And it's all, every member of the company, just talking about how they use AI in their workflow. So I think this is already becoming like the industry status quo of just like, "Yo, use AI, share how you use AI." We're all trying to figure this out together. But Toby, I think, kind of just made this shot across the bow for all tech company leaders to say like, "Yo, it is an expectation to be using AI, and if you don't use AI, you're out." Josh, when you read this memo, what'd you think?

• 41:53Speaker 02

  I was really excited about it. And I think it's one of those cases where he's just saying the quiet part out loud, because a lot of people don't want to scare other people who are looking for jobs. But he's basically saying, "We are going to stall hiring, and we are going to try to increase the leverage of our employees as much as possible." And I love the collective exploratory thing that he has, because I think that's the most valuable thing, going back to what we were talking about in the last segment. No one really knows how to maximally extract value from this, especially in narrow use cases like a particular company like Shopify. So collectively, getting everyone together, saying, "You must use this, and you must share your findings." I think- Is very productive. I forget the specific numbers. I think it was Shopify is getting 40% growth is, I believe, the number. And the expectation is that the employees can match

• 42:37Speaker 02

  that growth using this leverage tool. So he's expecting a 40% growth of production from individual employees just by leveraging these tools. I think it's super exciting. It is a little scary because this is a large company who is now saying, hey, we're not going to hire anyone unless you can prove that we really need them because AI is able to. And this is after they had a pretty large layoffs in the last 12 months. So it's, it is a new world order. It is a shift that is permanent. I don't see this changing the opposite direction anytime soon.

• 43:05David Hoffman?

  So let me just go ahead and look. There are 11,600 employees at Shopify. And so the CEO of a almost 12,000 employee company saying, hey, we are not hiring anyone unless you can prove that you cannot do that job with AI. I think that is pretty alarming from a jobs market's perspective, because that means that maybe, maybe you're not cutting anyone, but adding new jobs, I think, is not going to be a thing. It's going to start to diminish inside of Silicon Valley.

• 43:38Speaker 01

  Well, I also think a stat that goes underlooked in this memo is he's essentially asking every employee to 10x themselves. So he uses the example of, you know, you're all in your teams and then comes across that one golden employee that just 10x's the entire team, right? He says, well, now the tools that each of you use and have access to has the ability to 10x yourself. So, you know, it's not just a matter of like, we're not going to hire employees and you guys maintain the 20 to 40%, which is as he states here. I think he kind of wants people to like 10x the ability of each of themselves, which would mean, you know, what does a potential 100,000 plus version of Shopify

• 44:22Speaker 01

  look like, right? It's a $100 billion company as it sits right now. And Toby is, I think, due to his history, definitely been one to push with these frontier technologies. Do you guys remember when NFTs were the craze? Shopify was one of the first ones to test it out, you know, token gated access to A, B, and C service, right? And, you know, as necessary as they were, they canned it. But like, I feel like

• 44:46David Hoffman?

  God damn it.

• 44:47Speaker 01

  I feel like AI is just a little bit of a different point. And the second thing to your, that I want to kind of emphasize Josh's earlier point, which is

• 44:58Speaker 01

  There's a slight nuance here because people have been afraid that AI is going to replace them. But he is saying the quiet part out loud, but doing it himself. And I think CEOs are actually, and this is the subtle nuance here, going to push AI onto their employees and like oust their employees first. So it's like a human enforced displacement versus AI, you know, creating this like tool that just replaces you entirely. And then you just kind of like slot it in, do you know what I mean? We'll see approaches from both, you know, you've seen customer support agents replace people so far. But I think top down, kind of like pushing of AI tooling to replace individuals or stop hiring people is definitely going to become more of a trend.

• 45:39Speaker 02

  It feels like that rift is growing. And this is kind of the trend from the last segment, which was, are you going to use it for the Netflix version or for the productive output version? It's the same thing with Shopify and all these other companies where these are high leverage tools. Where if you are a person of high agency who uses them and leverages them, you will get very far ahead. But if you don't, you will be left behind by that amount of leverage. So it creates this very large gap that might not necessarily exist today, but will become increasingly large very quickly. And that probably causes a lot of concerns for people too, because now there are the people who leverage it. And now there are 10x employees and there are the people who do not, and now they are just one. And that's a much larger divide than someone who's like 5x versus 2x or whatever it may be.

• 46:21David Hoffman?

  We can even go back to the AI 2027 document that we were talking about earlier. Down at the bottom of the document, you know, towards the end of times, closer to 2027, this hypothetical AI lab, which you could just think of ChatGPT or OpenAI. They just didn't want to use OpenAI specifically. They even talk about all of the AI developers by the time that we approach the super intelligence explosion in 2027 that they think is coming, all of the AI developers that are working for the AI labs have all. Made themselves redundant and they're all kind of just there doing loose oversight without actually doing much of the work. And so even the, maybe even the AI labs will be the first people to actually fully AIify

• 47:06David Hoffman?

  their workforce because they are the ones who are intimately aware, they know how to use these things the best. And so maybe the canary in the coal mine is that actually AI labs start to use their own models as their own employees and all employees of like OpenAI or Anthropic or DeepSeek or whatever, they actually end up being AI managers more than actual like laborers themselves. That's kind of the model that AI 2027 gave in their document.

• 47:34Speaker 01

  Yeah, I would say that's the whole point, right? What is ASI? What is AGI? It is replicating human intelligence, but not just replicating it, just accelerating it completely. It just reminded me, OpenAI, didn't OpenAI tease their agents that they're releasing soon? And it goes through different tiers with the top one being $20,000 a month and it's like a PhD level employee that can come and work at your company. And then you had like a, something stupid like a 200 bucks a month and it's meant to be a high knowledge worker, which is basically like most people at like FAANG companies.

• 48:13David Hoffman?

  Most people, yeah.

• 48:14Speaker 01

  Yeah, most people. I definitely think that that's the case and like to go to the point of the AGI or AI 2027, they want to, one of the key themes is they want to use AI to improve itself iteratively over time. So it seems like that will be the main place that it gets displaced first.

• 48:33David Hoffman?

  Yeah. Okay, let's turn to education because this is something that came out this week. Claude, Anthropic just released Claude for education and this is basically trying to

• 48:46David Hoffman?

  AIify, you know, higher education up to the college level. And so this is, this is what it looks like. And they're basically, um, part of this, uh, part of this announcement comes a couple of different things. Uh, learning mode, a new Claude experience that guides students reasoning process rather than providing answers, helping develop critical thinking skills. Uh, university-wide Claude availability, full campus access agreements with Northern University, London School of Economics and Political Science, a few other colleges, making Claude available to students, academic partnerships and student programs. Uh, and so, uh, basically, uh, Claude is making a university as a product or a school as a product to help educate. And it's not just giving the, I think the important difference is if you go and type into chat GPT, hey, what's this answer? It'll give you the answer.

• 49:31David Hoffman?

  But here, uh, Claude education is, is working through how to arrive at the answer for people that are querying it. Uh, and then I think maybe one big headline that I think everyone should reflect on, uh, is this, uh, this one here from Foxnews.com. Texas private schools use of new AI tutor. That's Claude for education. Rockets student test scores to top 2% in the country. Uh, so a Texas private school started using Claude for education. Uh, and then all of a sudden their students start, starts testing at the top 2% in the country. Uh, and so this has massive implications. We already know the, uh, the university structure is already under threat.

• 50:11David Hoffman?

  Universities have been leaving, being the, going out of vogue for a while now, uh, because they are just entrenched ideological systems without any sort of like actual educational value. As students are taking in like quarter million dollars of debt to become like a, a speech therapist, uh, things like this. And so education has already been like needing a revamp. And now here we have AI, which for a use case that I think is just a match made in heaven. AI becoming your tutor, your teacher, uh, in this very streamlined way. Uh, Ajaz, what are your thoughts?

• 50:45Speaker 01

  I just don't see the point of universities within the next two decades going forwards. And I'm probably being generous. Two decades seems slow. I'm probably being super generous with that timeline, to be honest, right? If we're getting AGI in two years, why is that, you know, crazy to say? I, I, I think that there's a few points I want to make around this. Number one, the way it's packaged, this product called education, makes it seem like it's meant to aid lecturers. And to use your analogy earlier, David, you know, maybe employees become the coach of the AI, right? I don't think that's going to happen when it comes to education. I think it's going to become incredibly personalized. Heck, YouTube has already accelerated that massively. I've learned a ton of stuff that I wouldn't have normally learned via school, college, university,

• 51:29Speaker 01

  or any kind of prestigious degree that I would have ever attained. And I think personalized education just makes sense. Everyone is different, everyone learns and comes from a different background. Why would it make sense to, you know, use this arbitrary person? And sorry, I'm reducing professorship massively, but who has been, you know, honing their work and craft very, you know, commendably for decades, to then be expected to teach each individual in a certain or specific way? It just seems like AI is meant to scale this kind of thing. The Texas data point kind of proves that. But the second wider trend that I'm noticing here, David, is that I just don't think people are using AI effectively.

• 52:10Speaker 01

  Like, I would love to like contest and see what is so special about Claude education. I would bet you that it's the exact same model. It's just being crafted and curated in an educational circumstance. So behind the scenes, they've just given a long ass prompt, which says, "You are a professional university educator now, and you are going to specialize in economics 101. Please identify and analyze the London School of Economics syllabus for 2025, Term One, and create an agenda and schedule lecturers, et cetera, et cetera." Right? And then it's just being fed straight to the people there. So I, I put out this tweet recently. I don't know if you can.

• 52:52Speaker 01

  Pull it up, but I basically caught up with some friends and they were telling me how they spent four plus hours just this week alone prompting ChatGPT within a single chatroom to improve their life and how they learn. Now, if you look through this tweet, you know, when I asked why they were doing this, they said, so that I can tell me what job I should do next, you know? And this is in response to why me asking him why he put his entire career history into ChatGPT. And then another guy said he logs his entire set of notes from therapy and asks ChatGPT to give him life advice. I said, what do you mean? Like, what qualifies ChatGPT for this?

• 53:33Speaker 01

  But he goes, it doesn't matter who ChatGPT is, I just tell ChatGPT who to be. So I said, who did you ask him or it to be? And he goes, I simply said, ChatGPT, you are now Carl Jung, one of the forefathers of psychology, modern day psychology.

• 53:52David Hoffman?

  Carl Jung is Sigmund Freud's, Sigmund Freud. So he's hot Freud.

• 53:56Speaker 01

  That's a, that's a good way to describe him. And he basically said, here are all my therapist's notes. Also, you have the entire history of my chat log with you, talking about random arbitrary recipes and random Star Trek scenes that I missed, and then random hypothetical situations that, you know, I never ended up getting into, but I wanted your opinion on. Can you create an identity or personal personification profile of me? And then tell me what's wrong with me? How can I improve? Should I have broken up with that girlfriend or should I get back together with her? You know, it's this crazy thing. And then the third kind of like subtle point is like, we are just giving troves of our personal data to this thing, you know? I asked him like, how, how did you, on earth did you focus for two hours, giving career, all your career history?

• 54:41Speaker 01

  He said, oh, I just spoke to it. I just put it on voice chat and I spoke to it. You're just talking. I'm talking to another person there. And I think, you know, there's a dark side of it, which is like, you know, giving away all our data. We're gonna, you know, give away all control to this AI. But then there's the positive side, which is like, if you use this very well, you could end up having similar effects to that Texas school that shot to the top 2%, right? And it maybe had no business being there without AI, right? I don't know, Josh, do you have any thoughts on this?

• 55:10Speaker 02

  Yeah, I am in love with this headline because I have been at war with school for so long. It is such an abomination how people are educated. And it's, the AI segment is just the acceleration on the gasoline that has already been happening with self-education. In the case of high school, classrooms are pretty crappy. Most of them fall to the equality of the most disorderly student. It's a lot of disciplinary stuff. It's not very custom. The customization part, when you get to high or college, is the same thing. It's the same system for every single person. It's the same lectures, the same classes. Everyone learns differently. There is no reason why it shouldn't be hyper customized to you. And why you should not just have your own AI that understands your preferences, understands everything about you, and is able to deliver this information,

• 55:54Speaker 02

  educate you in the way that you most effectively learn. And I think the privacy thing is a concern. But even if the illusion of privacy is given, that's probably good enough because that's the reality we live in today. Where like all of your cookies are tracked, your online traffic is tracked, your GPS location on your cellphone is tracked. It's not very obvious. The average person wouldn't know that, but it still happens. So the privacy is a concern, but if you can get around that and give people these hyper personalized AIs, it is so much more effective than college. The only tricky part is if people do start self-educating using AIs and if they do get these hyper custom models, they lose the social elements of colleges and education and socializing in general,

• 56:35Speaker 02

  which contributes to this increasing trend of people just being a little more quirky socially.

• 56:40Speaker 01

  But we have the internet, right? We have a little digital avatar friend, right?

• 56:44Speaker 02

  But to me, it seems like there is absolutely no reason why it shouldn't be hyper customized. And it's reflected in this headline, which is like, the people who are using AI are now in the top 2% in the country. And that, this is very early AI. It will only ever get better from here and get more personalized as it understands your preferences. So I am all for it and very excited about it.

• 57:01Speaker 01

  Okay, so if we were to take the other side of that, Josh, some immediate thoughts that come to mind, and I'm curious what you both think, is, number one, alignment of these models need to be super important and governed very well. Can you imagine if the creator of a model came in with an agenda, you know, to sway the masses into a particular educational, political stance, for example, or just, you know, teaching basic math completely incorrectly? Obviously, that's a more extreme example. What happens when humans lose oversight, basically, right? And they stop fact-checking, which leads me to my second point, which is, do we just become incredibly lazy? It sounds super productive, right? It's like, hey, we're going to learn all these different things. But then do we just stop trying to pioneer?

• 57:44Speaker 01

  It goes back to the, what do we do when AGI manifests, right? Do we just become incredibly lazy and just said, ah, you know, let AI discover who God actually is and figure that out? You know, where do we end up?

• 57:57David Hoffman?

  Imagine verifying yourself without handing over personal data. No hacked databases, no unnecessary personal exposure for airdrops, and no AI bots ruining community governance. Meet SELF, the on-chain identity verification protocol built for privacy and control. SELF Protocol uses zero-knowledge proofs to confirm your identity safely. Users prove key details like age or citizenship without revealing sensitive personal information. SELF never stores your data. It only generates cryptographic proofs. Here's how it works in three steps. First, register and verify. Use the SELF app to scan your biometric passport's RFID chip. SELF verifies authenticity with zero-knowledge proofs. Each passport creates one unique identity. Second, you can share proofs privately. Third-party apps request identity proofs, like confirming you're over 18.

• 58:42David Hoffman?

  You can also link proofs securely to public wallets for airdrops or governance participation. And then last, secure verification. Apps validate your proofs instantly on-chain, like on Cello, or off-chain. Audited by ZK Security, the SELF app is live on iOS and Play Store. Visit self.xyz and follow SELF Protocol on X. The Arbitrum Portal is your one-stop hub to entering the Ethereum ecosystem. With over 800 apps, Arbitrum offers something for everyone. Dive into the epicenter of DeFi, where advanced trading, lending, and staking platforms are redefining how we interact with money. Explore Arbitrum's rapidly growing gaming hub, from immersed role-playing games, fast-paced fantasy MMOs, to casual luck-battle mobile games. Move assets effortlessly between chains

• 59:26David Hoffman?

  and access the ecosystem with ease via Arbitrum's expansive network of bridges and on-ramps. Step into Arbitrum's flourishing NFT and creator space, where artists, collectors, and social converge and support your favorite streamers all on-chain. Find new and trending apps and learn how to earn rewards across the Arbitrum ecosystem with limited-time campaigns from your favorite projects. Empower your future with Arbitrum. Visit portal.arbitrum.io to find out what's next on your web3 journey.

• 59:56David Hoffman?

  At the end of the day, school, the institution of school, public school at least, public school is basically glorified daycare just to get kids out of the hands of their parents so the parents can go to work and be productive. And then school just teaches you, public school just teaches you the basic modicum of knowledge, but really just puts you in a place where you won't get into trouble for like eight hours, six hours of the day. And that is so homogenous, because we are all going through this extremely homogenous school program all across the country, all across the United States. Like, yeah, it differs from state to state, but like, not really. And I think what we will see with education is the same thing as what YouTube or Instagram Reels did to mainstream media,

• 1:00:40David Hoffman?

  where there were just like seven news channels, and then that exploded into 70,000 YouTubers, all giving their takes on the news, some terrible, some great, some way better than mainstream media. But it's really, you know, choose your own adventure. And now we are going to see that fragment even further into just every single student has their school. It's one school per student. I don't know if you guys went to public school, but I remember class sizes were a huge issue in high school, where they were packing in 35 kids into a classroom, and it always got bigger every single year. And this is the exact opposite of that, where you get one hyper-personalized teacher, professor, who's the best at their job

• 1:01:23David Hoffman?

  to teach you exactly what you want to learn. Have you guys ever asked ChatGPT, "Generate an image of me based on everything that you know about me"? You guys have done this?

• 1:01:35Speaker 01

  Yeah.

• 1:01:35David Hoffman?

  Okay, so it comes, it is, you first you forget, you first realize how much data you have given ChatGPT about yourself in ways that like you have not given any data to any other application before. Because like it's truly friendly, it's a friendly, ChatGPT especially is very supportive. And so you, anyone who's using ChatGPT, go open it up, ask, generate an image of me based on everything that you know about me, and it'll generate an image of you, and you will realize how much it knows about you and your style. And think about how high fidelity that relationship that it has with you, and now, now, now it's your professor, right? It already knows you, all of your interests, your style and your quirks.

• 1:02:19David Hoffman?

  And now it's, now it's your school system to carry you or your child from like six years to 16 years old. I think it's, I think it's really cool. I'm not really in my learning years anymore, but nonetheless, I'm still looking forward to like learning some university level education from ChatGPT.

• 1:02:38Speaker 01

  I just go rejected.

• 1:02:39David Hoffman?

  Or clock, excuse me.

• 1:02:40Speaker 01

  I just go rejected, David.

• 1:02:41David Hoffman?

  You just go rejected?

• 1:02:42Speaker 01

  I don't have any personal details about your appearance or characteristics, so I can't create a tailored image of you. If you'd like, you can provide me with some descriptive details, blah, blah, blah, blah, blah. What do you mean? I've given like my entire life to this thing.

• 1:02:53David Hoffman?

  You just, you just force it to make an image. Just say, make your best guess.

• 1:02:57Speaker 01

  Okay. I'm gonna give it a go.

• 1:02:59David Hoffman?

  We'll see how it goes. I have seen this come out wrong, where like my, you know, generic looking white friend came out as like a black overweight female.

• 1:03:06Speaker 01

  Nice.

• 1:03:07David Hoffman?

  And I was like, wow, that's, that's different.

• 1:03:09Speaker 02

  Mine was startling to the point where it hadn't, the actual view out my window was correct. It knew what I was looking at and everything. I was like, wait, this, did I ever tell you this?

• 1:03:15Speaker 01

  Wait, wait, wait, wait, what do you mean? What do you mean? And I had the William Obu? I don't want to dox you, but yeah.

• 1:03:21Speaker 02

  Yeah, no, it had, it had the bridge in the background of my image, and I was like, wait, that's the exact view that's outside the window.

• 1:03:26David Hoffman?

  Did you ever ask it about moving?

• 1:03:28Speaker 02

  Well, we must have spoken about it, because it didn't just learn this on its own. So we were talking about something, moving, whatever I was looking at, trying to figure out the name of whatever street it was, but it knew, and it was very accurate. It had a little drone in the background and a camera and a microphone, and I was like, wait, this is weird, that's like kind of like my room.

• 1:03:43David Hoffman?

  Okay, you guys want to see mine, my image?

• 1:03:45Speaker 02

  Oh, please, I'd love to.

• 1:03:47David Hoffman?

  All right, here we go. Here's me.

• 1:03:51David Hoffman?

  For the listener, we have, we have a guy who's definitely on the Brooklyn side of New York with the New York skyline in the background.

• 1:04:00Speaker 01

  He's holding an espresso.

• 1:04:01David Hoffman?

  I have a rock climbing rope slung around my shoulder, an Apple Watch, a climbing helmet, and an espresso martini because it knows that I ask how to make cocktails pretty frequently. This is pretty good. It's pretty, but also it also works through its logic as well. It just tells you its logic and its workflow. It's like, pretty cool.

• 1:04:20Speaker 01

  Wait, can I, can I share mine? It actually surprisingly nailed it. One second.

• 1:04:25David Hoffman?

  Yeah, send me the picture and I'll put it on the screen.

• 1:04:28Speaker 01

  Here we go.

• 1:04:31Speaker 01

  Nail it!

• 1:04:33David Hoffman?

  That's funny because I think we both look like generalized AI engineers.

• 1:04:38Speaker 01

  Yeah, yeah. For some reason, I look entirely like, you know, and even nerdier David Hoffman. That's crazy.

• 1:04:44David Hoffman?

  Much, much nerdier. Yeah. You are not the attractive Indian man that we see on the screen.

• 1:04:50Speaker 01

  Maybe potentially a tech billionaire? I can't quite tell. Like, I know he has, this guy has at least 12 hoodies. Yeah, wow.

• 1:04:58David Hoffman?

  All right, let's bring up the rear of this, this AI rollup as we close this out. The last big one is the news of the week is A16Z's, not AI16Z, the actual, a real A16Z, raising a $20 billion mega fund to invest in AI startups in the United States. So that's pretty damn big. I mean, this is A16Z's job. It's not any surprise that they're raising $20 billion, but I think it just goes to show how much juice there is left to grow out the AI industry. Jaws, when you saw this headline, $20 billion being raised, not yet raised, but being raised by A16Z to invest in the United States AI, what were your first thoughts?

• 1:05:37Speaker 01

  So I had an initial kind of like gasp, which I was like, that's a humongous number for a single fund. But then I was like, oh yeah, no, that that that probably makes total sense, right? Like, it wasn't too long ago where, you know, Sam Altman was prospectively, you know, trying to raise a trillion dollars or whatever crazy figure, was it eight trillion, seven trillion dollars, something crazy like that? Yeah, so I don't think this is, you know, out of turn, especially given like some of the valuations that are coming. Do you remember, I think like Curse, or literally last week, got like a $10 billion valuation just based off its recent round? So I think, you know, these numbers kind of make sense. I I'm kind of curious where it's going into, right?

• 1:06:18Speaker 01

  And I think what's most interesting about this particular announcement is they didn't just, they weren't kind of lazy about it. They weren't kind of just like, yeah, we're gonna dump a bunch of this into like compute and data center companies and la la la, create a model. They were like, no, we want to own the app layer of whatever this behemoth of technology ends up becoming. And they use the term vibe creation platform. And so for those of you who are wondering what the hell that is, think of Curse, which is a vibe coding platform. But instead, it's a platform that can hit almost anything that you want to do, right? Not necessarily coding, it could be, I want to become the next biggest music artist

• 1:07:01Speaker 01

  and perform on my concerts all around the world. Well, to start off with, I need to be able to create some music, right? So what if I go on this vibe creation music platform and I can, you know, become the next superstar, right? And what's the difference between like, you know, just like this and maybe, I don't know, Apple's Garage Studio, where you can kind of like mix beats, is one, it uses AI, so you can just like prompt and type in the kind of flow and you can edit your music and all that kind of stuff. But then you could, it also connects to your socials directly. It also ties and brings in all your friends to listen to it in a really easy, simple manner. What they're trying to do is create this unanimous platform, right? Or they're trying to fund founders that are building this unanimous platform to connect different things. It doesn't have to be music, it could be movie editing or whatever that might be.

• 1:07:47Speaker 01

  And why I found this so interesting, and I'm curious to hear whether you two have the same take. Obviously, we've spent like a lot of our time in the Web three world, right? And the whole promise of the Web three world was to kind of connect all these different intermediaries and entities to create some kind of unanimous experience. It kind of sounds like what they're trying to do. And I am super happy to see this investment happen at the App Player. I mean, Josh, what do you, what do you think?

• 1:08:16Speaker 02

  I'm just so excited to see a lot of money going to something other than training. I think all of the money that's been raised is going to GPUs and to building larger clusters. And it's exciting to see smart people with a lot of money shifting away from that focus on cluster training to actual application layer. It goes back kind of to what we were talking about earlier, where there is just this single text window and you're supposed to extract this value. And they're slowly narrowing those use cases. And I think that's the most valuable thing that you can do is just teach people how to get value out of this. So using your example, he says, if you want to become a professional musician, this is the narrow box that we have built that is optimized for allowing you to do that thing. And so far, those solutions don't exist. It's still very broad. So investing $20 billion into creating narrow but hyper useful AI for whatever you might

• 1:09:00Speaker 02

  be interested in is so, so exciting. So I'm very pumped about this one.

• 1:09:03David Hoffman?

  I remember reading a tweet thread last week about how this one, like a mall, a commercial real estate investor, had just revamped his entire process with architects that he would work with for his whole like mall revitalization process. So the idea here is like he would buy up a cheap distressed mall. He would work with an architect to plan a new design, revitalize it, and then increase the value of the mall that he bought, he invested in. And the architect's design process, he said, would be so much back and forth about how to make a good looking mall, like six to nine months of back and forth in labor with his architect. And he was able to take a picture of this mall, prompt ChatGPT to get something that

• 1:09:48David Hoffman?

  was 90% of the way there. And that was the new starting point that he had with this architect to go and he said it just cut off like eight out of the nine months of timeframe for that one development process and saved like tons of man hours. And the whole thing That I was thinking about when I was reading this is he is doing this inside of ChatGPT, which is unoptimized to do this job, but it is still the right brain to do it. So like, just like Josh has been harping on this entire episode, where this form factor for how we engage with ChatGPT is so constricting, because it's just this text box that looks like Google. But nonetheless, it is doing incredible work. It doesn't take that much more effort and labor from a startup

• 1:10:29David Hoffman?

  to make a architect and design optimized interface for ChatGPT. Just another wrapper of ChatGPT for a more narrow, but specialized application on this thing. And you could apply that across every single industry under the sun. And we just, there's just so much investment in L1. I see the same pattern here between the L1 investment or the FAT protocol investment from the crypto space, where people want to invest in the layer ones. They want to invest in Ethereum, they want to invest in Solana. Do they want to build a layer two? Or do they just want to build another layer one? They just want to invest in the blockchains and they don't want to invest in the apps. And that is the issue plaguing the crypto space. And right now, we have plenty of layer ones. We have Anthropic, we have OpenAI, we have DeepSeek, we have all of these different.

• 1:11:15David Hoffman?

  Goddamn, LG, the television company, came out with an AI model layer one. And right now, I think what Josh is saying is like, yo, we need the app layer of all these layer ones. We need all these layer ones to become more expressive, more opinionated, more directed at specific outcomes. And I think that's what Josh is excited about with A16C, actually building out the app layer and not just funneling more money into just making ChatGPT that much better in a very generalized, broad, diffuse sense.

• 1:11:46Speaker 02

  And it goes back to the benchmarks too, is now you have these like predefined benchmarks across these categories that you can actively compare, because it is a more narrow band of scope. So you can see who can actually create the best architecture drawings. And that is a clear benchmark that you can set once you have these narrow focused purposes. So that's another way to solve the benchmarking thing is like, okay, you could have your crazy good models, but narrow it down into something that's actually useful for me. And then we could compare how that helps me personally.

• 1:12:11David Hoffman?

  I love how we just went full circle with that. Well, I love that. I think that's a great way to end it. Josh, Jaws, thank you for helping the Bankless Nation go through the weekly news this week. It was a big week. Next week will be even bigger and we'll be back in seven days to bring you the news in the AI space. Josh, Jaws, thank you guys so much. Also, thanks for having me. Bankless Nation, you guys know the deal. We didn't even talk about crypto, so I need to find a new sign off there. I don't know if there's any disclaimers needed because crypto is risky, but we didn't talk about it. So I guess this is the frontier, software everyone, but we are glad you are with us on the Bankless journey. Thanks a lot.