Claude Makes Bio Discovery in Wet Lab, OpenAI Solves 100 Math Problems

SPEAKER_1 (0:00)

Welcome to the podcast. Today we want to talk about one of the newest breakthroughs that Anthropic has had in biology. There's a bunch of drama, of course, with Anthropic and Dario saying we need to pace the frontier. And at the same time, they were saying they were opening a wet lab in San Francisco. After 21 hours of the wet lab being open, it has already made its first discovery, and Dario has been talking about it, which has led to a lot of excitement, but also backlash from people concerned that Anthropic is going to slow down the development of a lot of great scientific things. There's people worried that they're going to, you know, create the next COVID. So there's all sorts of sizes of the conversation we're going to be breaking down what's going on, what they're discovering in biology, as I think this is one of our most advanced movements when you have a company like Anthropic putting the full force of their research power behind AI and science. This is going to get interesting today on the show. Before we talk about that, I wanted to mention if you use Claude and you use something like Claude Cowork, which is an amazing tool for getting stuff done on your computer, and you're frustrated that Claude Cowork doesn't have the ability to generate images, audio, or video, I would love for you to check out my own startup, which is called AI Box. We created a connector that lets all of the other AI tools be generated inside of Claude. So I recently had an experience where I wanted to get a ton of images generated for a project I was working on, and I had this connector inside of Claude, and I told it the types of images I was wanting, but they were all OpenAI images to be generated. Claude was able to go generate over 600 images and put them all into the project, and the traffic on that site consequently is now skyrocketing, and it's up like 10x in the last three months. So overall, there's so many incredible capabilities. Claude doesn't have them all. If you want to add all of the audio, image, and video generation capabilities into the projects that you're working on with Claude or with any other AI tool like Muse or ChatGBT work, you can get the AI Box MCP connector. There's a link in the description.

It's only $8.99 to get started, so try to keep this affordable for everyone. Link in the description. Get all the AI models to work together. Okay, let's talk about the big announcement that Dario made. It was in response to Anthropic coming out with a tweet. This is just two hours ago where they said, Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriographs based on the enzyme's gene. Beside the enzyme's gene sits a long array of repeated DNA, a structure that looks somewhat similar to CRISPR Okay, so Dario came on and gave kind of a breakdown of what's actually happening over on X.

He said that, you know, they're announcing the Claude-led discovery. So obviously, it's not completely just Claude running all by itself. Claude-led meaning, like, Anthropica's funding this. There are real scientists involved. But what they're doing is having a bunch of Claude agents run a bunch of experiments to try to figure stuff out. He said that, he said, We suspect this could represent a new gene-editing mechanism. Its precise function, biotechnological utility, if any, or level of significance, is not yet clear. So, you know, they discovered something new. They're like, look, we found something new that no one really knew about. We don't know if this is important, if it means anything. This came after, I think, 21 hours of their of their white lab being online. So it's not like, you know, there's a lot of research that still needs to happen. But at a minimum, I mean, this is really exciting to have sped up. Dario said, the work was done mostly, though not entirely, by Claude. Life Sciences team suggested a broad area of research. Claude read through the literature and a bunch of genome data and discovered something interesting. Then Claude proposed experiments to verify the discovery, and our team carried them out. Okay, so right there, I think a lot of people are concerned. And by the way, there are actual companies working on, like basically getting robots attached to AI models, and the robots are, you know, mixing the chemicals in the vials and doing all the experiments themselves. And some people are freaked out about that. In this case, you know, there's a research team and they're like, hey, there's probably some things to discover in this area. Perhaps those researchers have already done research in those areas and know that there's a lot going on there. So they point Claude in the right direction, go look at these things. I mean, it's just like you or I, you know, going into a domain that we understand deeply and we're asking Claude or ChadGPD to help us with and do some research so we can figure stuff out. It's the same concept. So like it's quote unquote Claude led, Claude's the brain, but like a lot of these models are doing the research for everything nowadays, but whatever. I digress. I do think it's cool. Anyways, Claude goes, does some research, says, hey, look, I think I might have found some things, go do these experiments. And they have humans going and doing the experiments. Okay. Dario said, it's easy to dismiss this as a one-off or curiosity, but we've repeatedly seen a pattern where AI performance in new intellectual domains goes from weak to superhuman in a matter of a few years. In 2023, models struggled to do math at the level of an average high school student. In 2024, they started to do well at math competitions for the best high schoolers in the country. In 2025, they started to solve minor open problems. In early 2026, more significant open problems. And in late 2026, they're beginning to solve the top few open problems in all of mathematics. We believe AI for biology is on a similar exponential trend. What they're specifically referring to here is that we actually have a big kind of thing that came out from OpenAI, where they've actually put together like a math advisory board, they're calling it, at IA or at IAS, after they had OpenAI apparently resolve about 100 open problems. So, you know, OpenAI, there's a lot of drama when they solve a big famous math problem that other mathematicians have been studying or working on, and people, you know, speculated that they were getting people's chat conversations, and OpenAI denies it. Regardless, a lot of mathematicians right now are, some of them are saying, hey, you know, AI is ruining the field of math. You know, we can't intellectually work on stuff. I think there was nine mathematicians that were, you know, recently there's a whole, I think 25 field medalist winners signed an open letter earlier this month. They're warning that AI labs are, quote, threatening mathematicians' intellectual work, because AI could do a lot of the work. So there's obviously drama here. But in response to all of this, OpenAI has formed kind of a math advisory group where they're like, hey, look, guys, we are getting a lot of math done. We're solving a lot of problems. We would love to put them back to actual mathematicians to go verify and check what's going on. This group that they have in particular, they put together, they said, like, the group isn't allowed to tell OpenAI to slow down, or they're not allowed to do as much research. It's just an advisory group that's gonna help oversee the checking of a lot of these things. So apparently, OpenAI says, look, you know, there's one big famous one. We have a hundred other math problems that are big famous math problems that we've solved. And a lot of these, like, there's, you know, a million dollar prizes. So a lot of mathematicians are excited and are working on this. And some people are, you know, complaining, like I mentioned, 25 people signed that open letter, where they're like, look, you're ruining mathematicians' ability to do, like, complex stuff. But at the end of the day, these AI tools are just calculators. And so just because they figured out how to do something and they're a really complex calculator, in my opinion, that just means mathematicians should be able to solve more problems. And if these are problems that have been around for a hundred years unsolved, gee, wouldn't it be awesome to just solve them and move on with our life? Like, I don't really care if there's a, quote unquote, field of mathematics where it's people's job to solve these problems and figure, and spend a lifetime figuring out, trying to win a million dollar prize. If we found the answer to a problem we didn't know the answer to, just publish it and get it over with, now implement it into whatever it's useful for, and let's move on with society. So that's my opinion, and I'm sure it triggers people or whatever.