2026 - 02 February

Anthropic and the DoW

A more dramatic by Dean Ball (currently with the White House Office of Science and Technology Policy), and not an optimistic one.

The usually-sharp Ben Thompson writes, in my opinion, a very strange take on it in Anthropic and Alignment Two related good essays here as an overview, and as an oblique one which doesn’t use the word AI.

Doing physics with Ralph loop

Teaching an AI to Do Physics by Wiping Its Memory

This was a physics professor who discovered the power of Ralph Looping an AI agent, slowly letting it build on it’s failures and make new notes, until it does something most people assumed was well beyond possible.

AI can now do research-level calculations—unreliably and with the right setup. That’s genuinely new. But verification remains hard. The gap between “got the right answer this time” and “reliably gets the right answer” is still enormous.

Reminded me of the Brian Scalabine quote: At least at math and calculation, Claude is closer to Terrance Tao than you are to Claude.