Google Research's sleep framework lets LLMs consolidate memory offline, retaining up to 10M tokens
iruletheworldmo · x · 2026-11-01
Google Research introduced the sleep framework, a continual learning paradigm giving LLMs a periodic lifecycle to fight catastrophic forgetting and static knowledge cutoffs.
- Inspired by how brains consolidate memory during rest: models alternate between active interaction (wake) and offline memory consolidation (sleep), turning fragile short-term context into stable long-term parameters
- Results: continuous acquisition of new facts, retention of long-context knowledge up to 10M tokens, and improved complex reasoning
- Motivation: continual parameter updates cause catastrophic forgetting, while ICL alone forgets once the session ends
By Ali Behrouz and Vahab Mirrokni.
Related event: Google's Sleep Framework Enables Continual Learning for LLMs(2 posts)→
More from Research
- Humanoid motion-capture dataset bones-studio/seed trends on Hugging Face — bones-studio · 2026-11-01
- Terence Tao Slams AI Math as 'Strip Mining': Navier-Stokes Work Now 'In Disarray' — rohanpaul_ai · 2026-11-01
- ResGate cuts agent resource-hijacking attack success from 84.1% to 23.6% — jiqizhixin · 2026-11-01
- Two worlds: France's star mathematician shuns AI while anonymous peers push further with it — IgorCarron · 2026-11-01
- Stanford-Elsevier top 2%: Swiss AI researchers lead with 12.0 citations per paper — NegotiationOk7535 · 2026-11-01
- UIUC, Meta, Amazon & DeepMind release 135-page survey charting agentic reasoning for LLMs — mdancho84 · 2026-11-01