Research map / Large language models / Agents, code and math
Can Language Models Discover Scaling Laws?
Haowei Lin, Haotian Ye, Wenzheng Feng, Quzhe Huang, Yujun Li, Hubert Lim, Zhengrui Li, Xiangyu Wang, Jianzhu Ma, Yitao Liang, James Zou
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 |
|---|---|---|---|
| use SLDAgent evolutionary co-optimization of law and optimizer instead of native agent scaffold | GPT-5 / Codex · SLDBench avg | R^2: 0.550 -> 0.748 | Improved |
| use SLDAgent instead of Gemini-CLI | Gemini-2.5-Flash · SLDBench avg | R^2: 0.077 -> 0.506 | Improved |
| use SLDAgent instead of Claude Code | Claude-Haiku-4.5 · SLDBench avg | R^2: 0.419 -> 0.519 | Improved |
| use SLDAgent instead of Gemini-CLI | Gemini-3-Pro-Preview · SLDBench avg | R^2: 0.382 -> 0.636 | Improved |
| use SLDAgent instead of Codex | o4-mini · SLDBench avg | R^2: 0.349 -> 0.657 | Improved |
| SLDAgent vs human-derived law | SLDAgent GPT-5 · lr&bsz | R^2: -0.076 -> 0.604 | Improved |
| SLDAgent vs human-derived law | SLDAgent GPT-5 · u_shape | R^2: -1.000 -> -0.305 | Improved |
| SLDAgent vs Goose on u_shape and moe | SLDAgent GPT-5 · moe | R^2: 0.813 -> 0.773 | Hurt |
| SLDAgent with weak models on lr&bsz | SLDAgent (G-2.5-Flash, C-Haiku-4.5) · lr&bsz | R^2: -0.873 -> -0.871; -0.511 -> -0.657 | Mixed |
| SLDAgent-discovered moe law vs human law | SLDAgent · moe | R^2: 0.732 -> 0.891 | Improved |
| use SLDAgent law for fine-tuning model selection vs SubTuning | LLM selection · Wikitext | PearCorr: 90.3 -> 83.6 | Hurt |
| use SLDAgent law for LLM selection vs SubTuning | LLM selection · FLAN, Gigaword | PearCorr: 79.4 -> 84.6; 86.3 -> 94.8 | Improved |
Closest papers by meaning
- Observational Scaling Laws and the Predictability of Langauge Model Performance (NeurIPS 2024 · similarity 0.89)
- A Hitchhiker's Guide to Scaling Law Estimation (ICML 2025 · similarity 0.88)
- How Do Large Language Monkeys Get Their Power (Laws)? (ICML 2025 · similarity 0.87)
- LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling (NeurIPS 2026 · similarity 0.87)
- Language Modeling by Language Models (NeurIPS 2025 · similarity 0.87)
- AlphaZero Neural Scaling and Zipf's Law: a Tale of Board Games and Power Laws (NeurIPS 2025 · similarity 0.87)
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.
