Big Lab Researchers Claim Top AI Conferences Are Full of Overclaims
burny_tech · x · 2026-08-05
The AI academic community is debating the replication crisis in top conferences (ICLR/ICML/NeurIPS). A researcher pointed out that many big lab scientists now read almost zero papers, believing these venues are mainly full of overclaims and fraud.
To tackle this, the community proposed radical solutions: requiring all mathematical proofs to be formalized using tools like Lean, and mandating that all experimental results be automatically reproducible by an LLM directly from the PDF.
More from Fun
- Bank's AI customer support pretended to be human, staff denied it — Genzinvestor16180339 · 2026-08-26
- User generates 'Dolly Guitarton' via AI to honor Dolly Parton — charis_ai · 2026-08-26
- User Scolds AI Agent for 'Thinking Like a Human' — examachine · 2026-08-26
- User suggests Warhammer names for AI companies after exhausting LOTR — matt_slotnick · 2026-08-26
- Typo turns 'consider all angles' into 'angels', leaving Claude stuck thinking for 47 minutes — paulnovosad · 2026-08-26
- Mystery 'ox-alpha' model hype dissected: low-cost rollout playbook, likely Zhipu — lxfater · 2026-08-26