Tag
The paper introduces K-Bench 01, a benchmark for evaluating AI agents on real scientific requests, revealing that no model consistently meets the threshold for acceptable performance, with overclaiming as a common failure.
This tweet from Lior discusses agent performance in reinforcement learning with dynamic environments, while highlighting PatronusAI's $50M Series B funding led by GreenfieldVC for developing AI simulations and evaluations.
Harvey is experimenting with inference-time model routing and blended intelligence techniques to improve agent performance in legal tasks, highlighting both promise and risks.
An exploration of how AI agent memory systems often miss crucial cognitive processes like working memory, drawing parallels to anterograde amnesia, and offering design guidance for more effective solutions.
The article argues that the harness (the system around the model) is as important as the model itself for agent performance, citing evidence from various benchmarks and experiments.
A new research paper on δ-mem improves agent response quality by 7-32% when integrated with openclaw. The project is currently usable only with mlx and Qwen3:4b, but adapters for other models are expected.
Presents PREPING, a framework for constructing agent memory before any task-specific experience using proposer-guided synthetic practice, achieving competitive performance with reduced deployment costs.