NeurIPS desk-rejects 5/5/5-scored paper over one co-author's policy violation
AccBalanced · x · 2026-09-25
A researcher reports that Harness-1, their NeurIPS 2026 submission on RL for long-horizon search agents, received 5/5/5 from all three reviewers yet was desk rejected because one co-author violated reviewing policy.
- The author accepts sanctioning misconduct but questions punishing every co-author who had no knowledge of or control over the violation
- The work took 6 months; long-horizon search agent RL is expensive to run
- Resubmitting to ICLR isn't trivial: some frontier LLM baselines are no longer current, so experiments may need redoing
- The author argues the policy deserves reconsideration.
More from Companies & People
- CVPR 2027 publicity co-chair plans meme leaderboards and a best-shitpost award — ziv_ravid · 2026-09-25
- PyTorch announces ExecuTorch Hackathon in San Francisco, Oct 17-18, with three device tracks — PyTorch · 2026-09-25
- Stripe publishes Europe startup and compute data as Nina Schick calls US compute edge 'not even close' — ZeroStateReflex · 2026-09-25
- Judea Pearl's 90th Birthday Symposium at UCLA Now Online — AlexTensor · 2026-09-25
- New bilingual LLM lesson, conceived by a human and designed with Claude Opus 5.5 — S_Conradi · 2026-09-25
- JPMorgan: Meta's Muse agent has the potential to dominate the AI space — alexandr_wang · 2026-09-25