Frontier labs go public and claim real science

Anthropic's S-1 is on the front page. The thread has no visible comments, so the filing itself is the news. Next to it, a post titled 'Three new discoveries in mathematics, computer science, and physics' links to a preprint on OpenAI's GitHub, 'The Exponential Time Hypothesis is False'. One commenter said the title buries the lede. Another asked complexity theorists whether this moves their priors on P vs NP, while noting it doesn't imply P=NP.

The pattern: the big labs are now publishing results that outsiders have to take seriously, and they are also getting ready to answer to public-market investors. Sebastian Raschka's from-scratch reasoning-models course and Kauldron, a modular library for training ML models, show the builder side of the same story. People want to understand the machinery, not just rent it.

The open question is verification. A claim that big needs expert review before anyone should build on it, and the thread is mostly asking questions, not confirming anything.