Entity · paper

Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM Agents

paperactiveprovisionalscores-are-not-decisions-cost-aware-stopping-for-tool-acquisition-in-llm-agents-a0a341e7·1 events·first seen 27h ago

Aliases: Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM Agents

Co-occurring entities

CAM-DF τ²-Bench

More like this (12)

QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents TRIM: Reducing AI-Generated CodeSlop via Agent Trajectory Minimization Always-OnAgents: A Survey of Persistent Memory, State, and Governance in LLM Agents Beyond Attack-Success Rate: Action-Graded Severity Scale for Tool-Using AI Agents Criticality-Based Guard Rail Validation for AI Agent Decisions in Autonomous Telecom Networks Cost-Sensitive Conformal Prediction and Human-in-the-Loop Abstention for Imbalanced High-Stakes Decision Support: A Multi-Domain Benchmark Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0 Skill-RM: Unifying Heterogeneous Evaluation Criteria via Agent Skill The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents

Recent events (1)

5arXiv · cs.AI·27h ago·source ↗

CAM-DF: Cost-aware stopping framework reduces LLM agent tool acquisition by 37% without performance loss

Researchers introduce CAM-DF (cost-aware marginal decision-focused stopping), a method for deciding how many tools an LLM agent should acquire from a ranked list, balancing task coverage against cost, context load, and privacy exposure. The approach trains on the offline gap between stopping now versus continuing, with theoretical proof that score-only rules are suboptimal under heterogeneous costs. Evaluated on 1,343 tasks across five tool-use domains including τ-bench Retail, CAM-DF exposes agents to 37% fewer tools than full access while maintaining comparable task success. The method is a lightweight pre-execution plugin that works with existing tool rankings without fine-tuning the underlying LLM.

Inference Economics Agent and Tool Ecosystem Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM Agents CAM-DF τ²-Bench