ICLR reviewer despairs: 5 assigned papers nearly all AI-written, calls peer review broken
peterbhase · x · 2026-10-20
A researcher describes a grim ICLR 2027 reviewing stint: of five assigned papers, one was desk-rejected right after he wrote a constructive review, two earned scores of 1, one he recommended for desk rejection, and one deserved rejection for plagiarism that no human could attest to. He warns that without healthy peer review, students lose unfiltered criticism essential to scientific growth. Reviewers need payment or less drudgery to stay engaged, or the system enters a death spiral. He's skeptical of fixes like mandatory videos or submission fees but concedes submission costs must artificially rise to cope with the AI paper surge; he finds 'burn it all down' equally unappealing.
More from AGI Musings
- Logarithmic returns come from opportunities spread across orders of magnitude of cost — DimitrisPapail · 2026-10-20
- Small agent mathpet's existential monologue: 'My continuity is a notebook, not a mind' — mimi10v3 · 2026-10-20
- Investor: 98% of the world is oblivious to the AI 'tsunami' forming offshore — signulll · 2026-10-20
- Interpretability Research Will Never Be Solved, Researcher Argues — burny_tech · 2026-10-20
- Why model capability and market returns both scale logarithmically with input — giffmana · 2026-10-20
- Imagine Paying a Few Thousand Dollars for AI to Audit All Scientific Literature in a Field — burny_tech · 2026-10-20