Meta's PAMO lets a 4B personal agent lift recommendation hit rate to 47.9% using cross-app history
rohanpaul_ai · x · 2026-11-01
A Meta AI and UCSD paper formalizes Personal-Agent Mediated Recommendation: a platform ranks items with local data, while a personal LLM agent can rewrite the final list using user-authorized cross-platform history (e.g., Steam plus Nintendo/Xbox). The catch: platform rankings encode population evidence the agent can't see, so naive overrides often break good rankings.
Key contributions:
- MediateRec, a benchmark with scalable proxy cross-platform environments and a real cross-platform test;
- PAMO (Personal Attribution Mediation Optimization), which counterfactually masks cross-platform history to credit only agent answers that truly rely on it, and only when ranking accuracy doesn't drop;
- Theoretical guarantees that PAMO preserves cutoff-level advantage mass and is locally optimal.
On real Steam users, a 4B PAMO agent hit 47.9% top-10 hit rate vs 40.8% for the platform, beating standard RL on all 5 test sets. Takeaway: agents should defer by default and override only with specific evidence from user history.
More from coding & agent
- Karpathy's nanochat hits 59K stars: a full LLM training lab in one repo — techNmak · 2026-11-01
- Personal agents' third act: giving your dot ongoing responsibilities — craigsdennis · 2026-11-01
- Hands-on tutorial trains a decision model on Qwen3-0.6B from scratch — dl_weekly · 2026-11-01
- ui-ux-pro-max and 6 more repos giving AI coding tools real design taste — GCWebDesigner · 2026-11-01
- Hex ships stable release: local voice dictation for talking to AI agents on Mac, Windows, Linux — madeofAjala · 2026-11-01
- Personal AI agent race heats up — this prompt migrates your entire setup between agents — EXM7777 · 2026-11-01