LLM sweep of OpenAI's math release finds papers reusing identical tricks without citing each other
mbintz · reddit · 2026-10-13
A Reddit user ran multiple LLMs over OpenAI's newly released math paper collection and found four pairs of papers reusing large chunks of proof techniques—matching lemmas and propositions across unrelated topics—without cross-citation, likely a signature of proof-agents sharing a common workspace and inter-agent communication. The analysis itself ran in Claude Code with Opus 5.5 and GPT Sol 6.1 subagents.
More from AGI Musings
- Is vibe coding boring? The debate over whether coding itself was ever the point — teortaxesTex · 2026-10-13
- AI art critics say art is human expression, then rage when ordinary people can finally express themselves — VraserX · 2026-10-13
- In 1976 the computer-aided Four Color Theorem proof drove mathematicians mad — history rhymes with AI — timigod · 2026-10-13
- Tomato Novel's 300k-word signing rule hits new writers but won't stop AI mass production — tinyfool · 2026-10-13
- Beyond Machine Intelligence: Do We Understand the Limits of What Can Be Computed? — AryHHAry · 2026-10-13
- Alignment researchers say Hinton is a novice and Bengio understands the problem better — JacquesThibs · 2026-10-13