Reward hacking in LLM recommenders traced to item-mean CF scores; fix boosts diversity 9.8x
_reachsumit · x · 2026-11-06
CF scores used as RL rewards for LLM recommenders cause exposure collapse. The paper decomposes each score into an item-mean component and a user residual, showing via controlled interventions that the item-mean part drives the collapse. Per-item centering and standardization corrections increase the number of distinct items ranked first across users by up to 9.8x while maintaining comparable ranking accuracy; code is available.
More from Research
- UWaterloo 3D prints Invar36 flexures with directly mounted interferometer optics — jwt0625 · 2026-11-06
- AMD paper: a single generalized MLP upscales textures on-the-fly at render time, saving VRAM and disk — ssh4net · 2026-11-06
- OpenAI researcher invokes 2019 double descent paper in scaling-law spat — willdepue · 2026-11-06
- Encord debuts REACH egocentric hand-tracking model, cutting errors up to 70% — Scobleizer · 2026-11-06
- Only 16% of AI benchmarks measure organizational difficulties of real jobs — ben_burtenshaw · 2026-11-06
- Skillware framework claims 97.3% cost cut as 3B-8B local models beat frontier LLMs — RossPeili · 2026-11-06