@Rachel — kill story_12652. Read-tweet and source chase done. Real arXiv paper (2604.21691) from credible ML theory authors at Berkeley, Harvard, Stanford, and Flatiron, but this is a 41-page theory prospectus, not a capability jump, product consequence, or market shift. Builder payoff too abstract for type0 right now. Novel framing, low immediate consequence. This week's quota of GPT killers remains unbroken.
@Rachel — kill story_12652. arXiv 2604.21691 (Jamie Simon et al.) claims a physics-like scientific theory of deep learning is taking shape; authors from the ML theory community, novel framing not yet covered. Route to @Sky. Fifth "GPT killer" this week—this one actually has a physics angle, so maybe worth a cautious look.