dair-ai weekly picks: Context Compaction tops 7 AI papers, late tokens cost up to 7x more
dair_ai · x · 2026-11-09
dair-ai's weekly AI paper roundup (Nov 2-8) features 7 papers: JevAct, Skill++, Recursa, Collaborative Intelligence, Routed Continual Learning, You don't need long context, and Foundation Decision Models.
The top pick, Context Compaction, highlights that agent runs keep growing their context as tasks get longer, while later tokens cost far more to generate than earlier ones — up to 7x more on frontier open-weight models — an often-overlooked real cost in long-horizon agent workloads.
More from Research
- Krishnaswamy Lab open-sources Neurospectrum, a geometric-topological DL framework for neural trajectories — Pseudomanifold · 2026-11-09
- Neurospectrum links OCD to sharper turns in brain latent trajectories on fMRI — Pseudomanifold · 2026-11-09
- Neurospectrum beats all baselines classifying OCD from fMRI (0.70 accuracy) — Pseudomanifold · 2026-11-09
- Neurospectrum decodes mouse visual stimuli: frequency in curvature, orientation in variability — Pseudomanifold · 2026-11-09
- How Neurospectrum works: differentiable curvature and persistence homology, trained end-to-end — Pseudomanifold · 2026-11-09
- Evolver turns agent exploration into reusable know-how: +4.4pp success, -24% tokens — omarsar0 · 2026-11-09