Only 7 of the replicated ICML 2026 oral papers mostly held up, thread says
ChenhaoTan · x · 2026-07-24
A post about a replication project claims that, after reproducing ICML 2026 oral papers, only 7 of them mostly held up.
The quoted thread argues that as AI agents make long-horizon research work faster, the field should rethink its incentives and verify science rather than just produce more of it. The response praises tools that help reproduce papers because they preserve the quality and authority of research in a faster-moving cycle.
One comment also notes that the project was reportedly sponsored with $2.2 million, calling that a meaningful investment in research verification.
Related event: AI Agents Reproduce ICML Papers: Only 7 Out of 168 Fully Hold Up(5 posts)→
More from Research
- Qwen3.8-Flash-Next's open 180B weights hailed as a model organism for backdoor research — jd_pressman · 2026-10-27
- UK AISI papers: no good safeguards yet against misuse of biological AI tools — davidmanheim · 2026-10-27
- CMU's XYBench: LLMs answer queries literally and reinforce user misconceptions — valentina__py · 2026-10-27
- arXiv advice from T. Dietterich: wrong primary category will almost surely get your paper on hold — tdietterich · 2026-10-27
- arXiv has 4,000+ papers on hold amid submission surge; moderator shares how to avoid it — tdietterich · 2026-10-27
- Community-trained MOSS Sharpness Slider LoRA: 72 photos, 500 steps, +/-1 focus control — aoleg77 · 2026-10-27