Lightspeed's @buckymoore says the bar is now "way, way higher" for neolabs to succeed, but there's still an opportunity to exploit the "bureaucracy" that slows down larger labs: "Everyone talking about sample efficiency, everyone talking about continual learning — these are things that the large labs are very, very focused on as well." "But I also think there's a pretty valid argument to be made that the bureaucracy of these large labs is real." "In the same way that OpenAI was able to exploit Google's bureaucracy by scaling a transformer before [Google did], I think you could argue that such opportunities will exist in the future, with OpenAI and Anthropic being distracted by making the products that earn them all this money."