Meta's AIMS trains LLM recommenders to follow user requests over misleading history
_reachsumit · x · 2026-10-05
Meta researchers propose AIMS (Asymmetric Intervention-Guided Margin Supervision, arXiv 2610.02600) for instruction-guided generative recommendation. When interaction history conflicts with the current request, history events can override it. Naively converting per-event deletion effects into supervision fails: the most influential events aren't necessarily supportive, and removing a misleading event can raise a competitor's score even more. AIMS uses a frozen reference model to find deletions that improve both the target's score and its margin over a nearby competitor, converting those margins into ranking supervision while keeping full history as input.
More from Research
- Symmetric cryptanalysis took 6,677 research-years, 6.4x all lattice work combined — matthew_d_green · 2026-10-11
- 10th grader used free Muse to produce 3 verified math preprints in 8 hours — EastConsequence3792 · 2026-10-11
- ML interatomic potentials reveal fast Li+ conduction mechanism in glassy antiperovskites — TimothyDuignan · 2026-10-11
- Bittensor subnet Metanova uses competition-driven AI drug discovery with robot labs — const_reborn · 2026-10-11
- Lean4 proofs are not a silver bullet: consistency gaps, soundness bugs, and flawed benchmarks — elie · 2026-10-11
- Training inside the harness lifts Qwen3-14B from 22.2% to 54.8% on Spider 2.0-SQLite — omarsar0 · 2026-10-11