ICLR 2027 rebuttal catch-22: new experiments deemed "too much change," rebuttals dismissed as "AI slop"
SimpleObvious4048 · reddit · 2026-11-17
The author describes two losing scenarios in the ICLR 2027 rebuttal cycle:
- The "AI slop" dismissal: after running exactly the theoretical and empirical validation a reviewer requested, the full response was "the rebuttals are also clearly pure AI slop... I am keeping my score" — zero engagement with any new proof or number. The author notes ICLR 2027 explicitly permits LLM-assisted drafting as long as authors take responsibility for the content.
- The catch-22: on another paper, the team ran requested new experiments, transparently reported even unfavorable results, and rewrote the framing and abstract — only to be told "a paper that has changed this much is better evaluated as a new submission."
The author acknowledges reviewers are overloaded volunteers and some were fair this cycle, but finds the shifting bar exhausting: don't fix issues and you're rejected; fix them thoroughly and you're rejected for changing too much. The post invites others' experiences from this week's rebuttals.
Related event: ICLR authors lament no-win review process dominated by AI-like reviewers(2 posts)→
More from Fun
- Retro sci-fi AI image set "The Last Horizon" dazzles with crystal cities — Kyrannio · 2026-11-17
- Stunning close-up of Super Heavy's Raptor engines firing: 'There is beauty in humanity' — Kyrannio · 2026-11-17
- Anthropic employee's sarcasm at Aschenbrenner hints at internal discontent — teortaxesTex · 2026-11-17
- Joke: LLMs are now "language-image-audio-video-robotics-proteins-meta models" — burny_tech · 2026-11-17
- gwenshap: Parents make better AI agent users — kids train you for loopholes — josh_wills · 2026-11-17
- Grok bot phones a store and secures a $1,511 refund for its user — elonmusk · 2026-11-17