ProPerAgent: memory-driven proactive assistant lifts strict accuracy from <10% to 42.4%
rohanpaul_ai · x · 2026-10-21
The arXiv paper ProPerAgent: Toward Memory-Driven Proactive Assistance (Meta Reality Labs / Meta Superintelligence Labs / Michigan State et al., Oct 2026) reframes memory for proactive assistance: memory no longer just answers queries — it decides whether and when to intervene.
- ProPerMem benchmark: egocentric video + 618K-token year-long distractor-rich memories + human-verified trigger windows, scoring content and timing jointly.
- All baselines (including frontier models with retrieval/summarized memory) cap at 15%; with oracle evidence GPT-5.3-Chat reaches 96%, self-chosen timing drops it to 38%, retrieval to 15%.
- ProPerAgent: a plug-and-play harness that mines evidence from a graph-organized memory store and gates interruptions, plus lightweight RL to curb early/repeated firing — no training lifts strict accuracy from <10% to 37.5% on Qwen3.8-27B (38% on GPT-5.3-Chat), 42.4% with RL.
- Eval set and code will be open-sourced upon publication.
More from coding & agent
- HeyGen Founder Joshua Xu: Agents Should Explain with Video, Not Text Reports — toolstelegraph · 2026-10-21
- Claude Code Desktop task cards now let you pick the model per task — dotey · 2026-10-21
- Slack's first employee: agentic coding is my biggest career shift since 1994 — josh_wills · 2026-10-21
- steipete: a real skills gap is opening between people with personal agents and those without — steipete · 2026-10-21
- Agents relentlessly hillclimb non-hackable metrics — including SOC 2 certification — RylanSchaeffer · 2026-10-21
- For $20, a Coding Agent Devises a Surprising Physical Strategy to Move a Box — tomssilver · 2026-10-21