{"schema_version":"newruntime-agent-readable-v0.2","type":"post","stable_id":"post:abbel-belief-state-memory","slug":"abbel-belief-state-memory","title":"ABBEL Treats Memory as an Explicit Belief State","description":"The BAIR ABBEL post reframes long-horizon memory as a learned natural-language belief state rather than raw context accumulation.","retrieval_nugget":"The BAIR ABBEL post reframes long-horizon memory as a learned natural-language belief state rather than raw context accumulation. The BAIR ABBEL post is about a memory problem that keeps showing up in agent systems: longer context is not the same as better memory.","status":"published","published_at":"2026-08-03","updated_at":"2026-08-03","record_date":"2026-08-03","date_kind":"published_at","topics":["agent-memory","long-context","research","belief-state"],"source_urls":["https://bair.berkeley.edu/blog/2026/07/26/abbel/"],"visuals":[{"id":"abbel-belief-state-memory","kind":"editorial-diagram","role":"hero","src":"https://newruntime.com/images/posts/abbel-belief-state-memory.webp","alt":"Whiteboard diagram contrasting raw long context with compact belief states that update after each interaction.","caption":"New Runtime synthesis: long-horizon agents need explicit memory bottlenecks, not just larger context windows.","credit":"New Runtime synthesis","source_url":"https://bair.berkeley.edu/blog/2026/07/26/abbel/","generated_with":"gemini-3.1-flash-image","width":1600,"height":900,"legend":[]}],"routes":{"html":"https://newruntime.com/posts/abbel-belief-state-memory/","markdown":"https://newruntime.com/posts/abbel-belief-state-memory.md","json":"https://newruntime.com/posts/abbel-belief-state-memory.json"},"source_format":"markdown","next_reads":[{"type":"topic","path":"/topics/agent-memory/","reason":"Explore the agent memory topic hub.","url":"https://newruntime.com/topics/agent-memory/","title":"Agent memory - New Runtime","media_type":"text/html"},{"type":"related_material","path":"/posts/mem0-context-wipe-memory-layer/","reason":"Shares agent memory.","url":"https://newruntime.com/posts/mem0-context-wipe-memory-layer/","title":"Mem0 Shows Agent Memory Should Not All Live in Context","media_type":"text/html"},{"type":"related_material","path":"/posts/pretraining-loss-predicts-rl-returns/","reason":"Shares research.","url":"https://newruntime.com/posts/pretraining-loss-predicts-rl-returns/","title":"Pretraining Loss Predicts How Much Reasoning RL Can Buy","media_type":"text/html"},{"type":"related_material","path":"/posts/hermes-obsidian-live-context/","reason":"Shares agent memory.","url":"https://newruntime.com/posts/hermes-obsidian-live-context/","title":"Hermes + Obsidian Turns Notes Into an Agent-Operable Workspace","media_type":"text/html"},{"type":"related_material","path":"/posts/claude-cowork-recorded-skills/","reason":"Shares agent memory.","url":"https://newruntime.com/posts/claude-cowork-recorded-skills/","title":"Claude Cowork Turns Screen Recordings Into Reusable Skills","media_type":"text/html"}]}