Beyond achieving high task rewards, what two operational factors does JIT-Agent seek to preserve or reduce when selecting candidate harnesses?
0
1
Tags
Prep Sessions
Frontier Foundation Models, Capability Evaluation, and Just-In-Time Agent Harnesses @ University of Michigan - Ann Arbor
Ch.3 Adaptive Agent Harness Design - Frontier Foundation Models, Capability Evaluation, and Just-In-Time Agent Harnesses @ University of Michigan - Ann Arbor
Stage I Harness Customization and Preference Learning - Frontier Foundation Models, Capability Evaluation, and Just-In-Time Agent Harnesses @ University of Michigan - Ann Arbor
Dynamic Agent Scaffolding: Synthesis, Diagnostic Repair, and Evolutionary Optimization @ University of Michigan - Ann Arbor
Ch.2 Multi-Stage Harness Optimization - Dynamic Agent Scaffolding: Synthesis, Diagnostic Repair, and Evolutionary Optimization @ University of Michigan - Ann Arbor
Stage I: Task-Conditioned Customization and Preference Learning - Dynamic Agent Scaffolding: Synthesis, Diagnostic Repair, and Evolutionary Optimization @ University of Michigan - Ann Arbor
Related
Under what condition are candidate agent harnesses evaluated and compared in JIT-Agent?
In JIT-Agent, candidate agent harness selection is trained using preference learning.
Beyond achieving high task rewards, what two operational factors does JIT-Agent seek to preserve or reduce when selecting candidate harnesses?
Detail the full mathematical criteria required for candidate harness h+ to be strictly preferred over h- (h+ ≻_τ h-) during Stage I preference data construction in JIT-Agent.
Match each mathematical notation from JIT-Agent Stage I preference learning to its operational role.
Order the steps taken in JIT-Agent to establish and quantify harness preferences for preference dataset construction.
Based on the Pareto-efficiency condition defined in JIT-Agent, determine whether Harness A is strictly preferred over Harness B (h_A ≻_τ h_B) and explain the reason.