Meta FAIR Introduces AI Research Preference Models (RPMs): Ranking ML Experiments Before Spending GPU Hours
AITopTools Editorial TeamSeptember 6, 2026
What changed
Meta FAIR, Oxford and UCL introduced AI Research Preference Models, fixed AI judges that rank 15 proposed machine-learning experiments before any are run. The system selects one experiment, raising the average score on AIRS-Bench from 0.684 to 0.729 while reaching the baseline’s 24-hour result in about 15 hours.
What this means for you
Research teams could use this approach to choose promising experiments before spending substantial computing time, but the feed describes a research result rather than a public product. Its reported benefits are limited to the AIRS-Bench evaluation and may not apply to every type of experiment.