Meta's LION uses a sparse memory layer to fix evolution conflict in generative recommenders
_reachsumit · x · 2026-09-15
Meta researchers identify 'evolution conflict' in generative recommendation: heterogeneous user preference shifts optimized in a shared autoregressive parameter space let dominant behaviors crowd out underrepresented ones. They propose three principles (isolated memorization, reinforced evolution, scalable application) and build LION around a sparse Key-Value memory layer. Paper and code are public.
More from Research
- Microsoft and SJTU open-source Argus, an agent that ran 1,548 hours and solved a 20-year-old math problem — jiqizhixin · 2026-09-16
- Composing Continual Learning Mechanisms Boosts Long-Horizon Memorization in LMs — JohnsHopkins · 2026-09-16
- Gavel Elicits Native Skill Routing from a Frozen LLM's Hidden States — Tsinghua · 2026-09-16
- ScienceBuddy: Recursive-in-Recursive Self-Improvement for Scientific Agents — Shuhan Xue · 2026-09-16
- RCT: LLMs fail to significantly boost novices' wet-lab molecular biology success — shae_mcl · 2026-09-16
- Six years on, scvi-tools still widely used — outlasting foundation models and coding agents — anshulkundaje · 2026-09-16