Early Anthropic investor @mmurph reveals why most neolabs won’t make it: "There's 70+ neolabs now. We're in 10 of them. There's two strategies. One is an alternative universe, where the model is super strong. We did Skild for robotics. For robotics, it's how you train the model and what it does. It's dealing with things like gravity. They're just different physics. Chai Discovery for drug discovery. Chai has their own models to develop antibodies and things like that. Or Axiom Math, which is math-oriented models. Or look at Suno. Mikey and team built their own audio models to make that experience magical. They built the model to enable the service. Same with Wispr. We have one called Inception that does diffusion for text, which makes it super fast for things like coding. We've got another one that does continual learning. Then there's a bunch of models that are highly complementary to existing foundation models. We're not saying, 'This is going to go right at Anthropic and take revenue from them.' It's more, where are the adjacent opportunities that can benefit from a complementary model? Now, with 70 of these, and many of them raising a billion as an A and then two or three billion as a B. That's clearly not going to end well. But right now, with the desire for the technology and the talent, there's going to be some amazing outcomes. Like Fei-Fei's company, acquired for $8 billion before it really had any revenue traction. If frontier intelligence is the product, that product is incredibly valuable right now if you don't have it. So we will see some spectacular acquisitions and outcomes. And we're also gonna see a lot of big craters in the ground."