The AI-era publishing trilemma: verifying proofs vs. speed vs. avoiding getting scooped
jd_pressman · x · 2026-09-18
jdpressman lays out a dilemma for AI-assisted research: the properties we want are in tension — understanding the proofs you publish works against publishing quickly to preserve credit between model iterations, which works against keeping intermediate results private to avoid being scooped.
In the thread he mentions asking the model Astra to attempt Collatz for fun, noting that past a certain point such exploration simply isn't token-economical without building on other people's work. The exchange captures a structural tension as AI accelerates math research: attribution, verification rigor, and secrecy can no longer all be maximized at once.
More from AGI Musings
- 1988 sci-fi short story imagined an emergent Chinese Room that answered back — toptickcrypto · 2026-09-18
- OpenAI's Noam Brown: Air-gapping may not stop misaligned AI, safety bar must rise — basedjensen · 2026-09-18
- Martin Casado endorses the take that 'AGI' and 'ASI' are thought-terminating clichés — seanmcdonaldxyz · 2026-09-18
- Martin Casado on Noam Brown's air-gap warning: covert channels are an old, understood problem — tszzl · 2026-09-18
- People now use AI to message their own friends, and it may seed a startup — mobileraj · 2026-09-18
- Dwarkesh: Labs Will Hide Models During RSI; Delangue Calls Concentration the Biggest AI Risk — soumitrashukla9 · 2026-09-18