HALO benchmark: human-annotated eval finds over-proactivity is agents' top failure mode
2077AIDataFoundation · hf · 2026-11-02
Researchers from UC Davis, 2077AI, and USC introduce HALO, a human-annotated benchmark for evaluating proactive agents that must decide whether, when, and how far to assist without explicit requests. HALO-Arena covers 4 proactive need types (Dependency, Blockage, Risk, Opportunity) across 10 everyday domains and 50 personal environments. Key finding: over-proactivity is the dominant failure mode; useful initiative requires knowing when restraint is warranted.
More from Research
- J-Lens is just a first-order Taylor expansion, researcher points out — burny_tech · 2026-11-02
- Mollick: expert-level work benchmarks are brutally hard, and ARC-AGI, GDPval, METR are saturated — emollick · 2026-11-02
- Rebuilding a Protein Domain Distribution Pipeline After a Backup Failure — AryHHAry · 2026-11-02
- NVIDIA's Elastic Latent Recurrence lets one 4B model trade loop count for accuracy: 18→29 on AA Index — burny_tech · 2026-11-02
- IBM's STAIR retriever cuts invalid references 65× and hits 82.6% accuracy by using document structure — solyarisoftware · 2026-11-02
- ReLift Hits SOTA Dense 3D Tracking on WorldTrack With 2.7x Faster Pipeline — Technion · 2026-11-02