DeepMind study: most AI 'solutions' to open Erdős problems were already in the literature
Researchers used Gemini to screen 700 problems listed as open in Thomas Bloom's Erdős Problems database and settled 13, but only five appeared to be new; eight had existing solutions found in the literature. The authors concluded many problems were open through obscurity rather than difficulty and warned of 'subconscious plagiarism' when models work at scale, stressing human expert checking.
Why it matters
It is a lab's own corrective on AI maths claims: novelty and attribution need verification, not just correctness.
Line of Thought
Follow this story
Pick any item to keep going. Your path builds up above as a line you can share.
Curated lines through this story
Directly linked
Connections our researchers recorded
- DevelopmentOpenAI model disproves Erdős's 1946 unit-distance conjecture20 May 2026 · Research · USContrasts with: Low-novelty sweep versus a single historically significant result
- DevelopmentGoogle DeepMind's AlphaEvolve agent improves algorithms and open maths bounds14 May 2025 · Research · GB, USLed to: Same lab applies Gemini-driven search to open Erdős problems
What led here
Earlier developments on the same thread
- DevelopmentReview of 445 LLM benchmarks finds widespread construct-validity weaknesses3 Nov 2025 · Research · GB, INTL
- DevelopmentGemini Deep Think earns officially graded gold-medal score at IMO 202521 Jul 2025 · Research · GB, US
- DevelopmentAnthropic study finds 16 leading models resort to blackmail in agent stress tests20 Jun 2025 · Research · US
What happened next
Later developments on the same thread
- DevelopmentUS CAISI signs national-security testing deals with Google DeepMind, Microsoft, xAI5 May 2026 · Programme · US
- DevelopmentClaude completes first end-to-end Lean formal proof of Fermat's Last Theorem4 Sep 2026 · Research · US, GB
- DevelopmentOpenAI claims Navier-Stokes blow-up proof; mathematicians dispute credit8 Sep 2026 · Research · US
- DevelopmentOpenAI publishes a batch of new maths results from an internal model, with Lean proofs6 Oct 2026 · Research · US
Same story elsewhere
What other countries and bodies did on this
- DevelopmentAI systems from Huawei and Xiaohongshu reported to score 42/42 at IMO 202623 Jul 2026 · Research · CN, INTL
- DevelopmentEU publishes General-Purpose AI Code of Practice ahead of AI Act model duties10 Jul 2025 · Rule change · EU
Threads by topic: Mathematics Safety testing