Latest in AI
AI news and analysis
Source-linked AI news and explainers for curious readers. What changed, why it matters, and what to try next.
More Rules, Longer Lists? Benchmarking Self-Evolving Agents on a Hidden-Rule Curriculum
CLR‑Bench reveals that most self‑evolving agents struggle to infer hidden grouping rules, improving only modestly with explicit curriculum guidance.
Read full story
Beyond Plausible Prediction: Safety Sufficiency in Web World Models
The paper introduces the SafeWebWorld benchmark to assess whether web world models preserve safety‑c…
Circumventing the Computational Hardness of Computational Arbitrage
The paper shows that computational arbitrage in AI model marketplaces is NP‑hard in general, but bec…