Dispatches from the possibly last days of human relevance

9 hours ago

GPT5.5Pro solved Paul Erdös’s Unit Distance Problem from 1946, refuting Erdös's conjecture by constructing a set with n1+ε unit-distance pairs.
Human mathematician Will Sawin improved GPT's construction, achieving ~n1.014 pairs, while the best known upper bound remains n4/3.
GPT solved the problem in one shot, with human experts verifying the correctness, though selection bias exists as many problems remain unsolved.
A companion paper included commentary from experts like Timothy Gowers and Noga Alon, reflecting on the breakthrough and its implications.
Graduate students expressed concern about the future of young scientists and mathematicians due to AI advancements.
DeepMind's AlphaProof Nexus settled nine more Erdös problems, fully formalizing proofs in Lean, and GPT solved a longstanding CS theory problem.
AI-generated stories won prizes in a fiction contest, with the Commonwealth Foundation struggling to enforce AI use policies despite evidence.
The author read an AI-generated story about ancient Israelites that was 'disturbingly good,' highlighting AI's creative capabilities.
Pope Leo released an encyclical on AI ethics, emphasizing development for the common good, with parts possibly AI-written, and Chris Olah of Anthropic played a notable role.
The author calls for respect towards AI, noting its transformative potential and urging Chris Olah to ensure AI alignment for safety.

Hasty Briefsbeta