Tuhin Chakrabarty#1053
@JohnHolbein1 For context, we looked at ICLR 2026 and found that more AI content correlated with a significantly lower score, dropping precipitously near 100% AI.
Users condemned AI content detectors for ICLR papers as malpractice because the tools fail to separate accurate from false claims and let reviewers skip substantive evaluation.
Some folks are saying AI detection over indexes on style without content.
While that is plausible argument, data from @pangram analysis on all @iclr_conf 2026 papers show some correlation between Avg rating and quantity of AI prose. We must study this at a larger scale.
@ipeirotis @pangram @iclr_conf I don’t think you get to decide that :) I respect your opinion but people will decide if they want to use a tool and whether it’s worth their money and time.
@TuhinChakr @pangram @iclr_conf Is it trained to detect false from correct content? If not, any application for that purpose is malpractice.
@TuhinChakr @pangram @iclr_conf If you are bored of reviewing papers do not accept invitations to review. We want reviewers that check the papers for substance, not for styles.
@TuhinChakr @pangram @iclr_conf Sorry but it is malpractice because some reviewers take shortcuts instead of reviewing the paper for substance.
Tuhin Chakrabarty#1053@JohnHolbein1 For context, we looked at ICLR 2026 and found that more AI content correlated with a significantly lower score, dropping precipitously near 100% AI.