Researchers worry papers and meta-reviews are both getting AI-written
fredahshi · x · 2026-07-22
The quoted post says the author often ends up writing human-authored meta-reviews for papers that were themselves mostly AI-reviewed, and many may have also been partially AI-written.
They mention seeing a lot of 100% AI scores on Pangram and say the situation feels demoralizing, even if AI can clearly help in research.
The main point is not that AI use in research is inherently bad, but that the community should openly discuss what happens when:
- papers are increasingly AI-assisted
- reviews are also AI-assisted
- authors and reviewers are both using AI tools in hidden ways
The post frames this as a broader conversation about research norms rather than a solved problem.
More from AGI Musings
- AI benchmark-maxing ignores speed, cost and compute power — DevToD4 · 2026-07-22
- AI may need constitutional-style protections, says Dan Jeffries — Dan_Jeffries1 · 2026-07-22
- A 2030 prediction list imagines humanoid robots, bio-youth, and no work — rand_longevity · 2026-07-22
- A rethink of alignment: maybe the real problem is user alignment — ctjlewis · 2026-07-22
- A model may simply follow the wrong instructions, not fail alignment — ctjlewis · 2026-07-22
- An essay says Anthropic’s Claude may be moralizing users into dependence — theomitsa · 2026-07-22