Research map / Large language models / Agents, code and math
Predicting Emergent Tool Use in LLMs Before It Emerges: A Proxy Perspective
Bo-Wen Zhang, Yan Yan, Guang Liu, Xu-Cheng Yin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| weight proxy tasks by alignment and consistency (ablation) | Proxy-based prediction · T-eval ranking (1.8B) | reversed pairs of 15: 5/15 (w/o align) -> 0/15 | Improved |
| add consistency weighting (ablation) | Proxy-based prediction · T-eval ranking (1.8B) | reversed pairs of 15: 4/15 -> 0/15 | Improved |
| add alignment weighting (ablation) | Proxy-based prediction · CIBench ranking (7B) | reversed pairs of 15: 6/15 -> 1/15 | Improved |
| use proxy ensemble instead of validation perplexity | Tool-use prediction · T-eval ranking (1.8B) | reversed pairs of 15: 6/15 -> 0/15 | Improved |
| use proxy ensemble instead of scaling-law extrapolation | Tool-use prediction · T-eval ranking (7B) | reversed pairs of 15: 10/15 -> 1/15 | Improved |
| weight proxies instead of plain averaging (ablation) | Proxy-based prediction · T-eval ranking (1.8B) | reversed pairs of 15: 6/15 -> 0/15 | Improved |
| use Kendall rather than Pearson/Spearman for alignment (ablation) | Alignment metric · Baseline robustness (S=10, Ms=25) | baseline robustness: 0.500 -> 0.580 | Improved |
| use later checkpoint (0.5T -> 3.2T) for prediction of optimization schedules (ablation) | Proxy-based prediction · 6 training configs | reversed pairs: 5/15 -> 1/15 (1.8B T-eval) | Improved |
Closest papers by meaning
- Beyond Accuracy: A Cognitive Load Framework for Mapping the Capability Boundaries of Tool-use Agents (AAAI 2026 · similarity 0.90)
- Joint Optimization of Tool Creation and Use for Large Language Model Agents (NeurIPS 2026 · similarity 0.90)
- Understanding Emergent Abilities of Language Models from the Loss Perspective (NeurIPS 2024 · similarity 0.89)
- Observational Scaling Laws and the Predictability of Langauge Model Performance (NeurIPS 2024 · similarity 0.89)
- Training on the Test Task Confounds Evaluation and Emergence (ICLR 2025 · similarity 0.89)
- Large Language Models Think Too Fast To Explore Effectively (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
