Topic
harnesses
Latest news
- 16 OctAlphaDiana: measurement system for agent reasoning and interventions
AlphaDiana is a modular measurement system for evaluating agents that preserves harness-native control loops while enabling controlled experiments on tools, skills, and memory. The authors use it to compare multiple open-source models and harnesses and find that harnesses and component interventions do not consistently improve reasoning accuracy and must be evaluated alongside behavior and resource use.
Research · 1 source
Articles
No articles about harnesses yet.