ALTK-Evolve tops τ²-bench customer-service tasks without a frontier model
IBM Research credits learned memory, not a frontier model, for the τ²-bench customer-service result.
Stories per day · 30 days
5 stories, newest first · one card per story
Stories tagged #tau2-bench
POLAR: Ontology-Guided Risk Prevention for Tool-Calling LLM Agents
POLAR prunes risky tool calls using ontology-based reversibility scores, with mixed tau2-bench gains.
DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks
DAEDALUS bootstraps reusable agent memory from self-generated practice, lifting success by up to 15.9 points.
Just-in-Time Memory: Learning to Curate Task-Adaptive Memory for LLM Agents
Just-in-Time Memory curates raw agent trajectories at read time and beats write-time memory on three benchmarks.
EDGEGEN: Improving Tool-Calling Agents Beyond Happy Paths with Synthetic Edge Case Generation
EdgeGen synthesizes database-grounded edge-case tasks that improve tool-calling agents without human annotation.