Research map / Large language models / Language, safety and interpretability
Sculptor: Empowering LLMs with Cognitive Agency via Active Context Management
Mo Li, L.H. Xu, Qitai Tan, Long Ma, Flood Sung, Ting Cao, Yunxin Liu
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 |
|---|---|---|---|
| add ACM tools with task-specific prompts | Claude-4-Sonnet · NeedleBench-M-RS | accuracy: 67.0 -> 94.0 | Improved |
| add ACM tools with task-specific prompts | GPT-4.1 · NeedleBench-M-RS | accuracy: 48.0 -> 71.0 | Improved |
| add ACM tools with task-specific prompts | DeepSeek-V3 · NeedleBench-M-RS | accuracy: 50.0 -> 58.0 | Improved |
| add ACM tools with task-specific prompts | Claude-4-Sonnet · PI-LLM | accuracy: 84.04 -> 89.94 | Improved |
| add ACM tools with task-specific prompts | DeepSeek-V3 · PI-LLM | accuracy: 57.39 -> 57.27 | No clear effect |
| add ACM tools with task-specific prompts | Claude-4-Sonnet · PI-LLM short contexts (4/1K) | accuracy: 99.13 -> 90.43 | Hurt |
| add ACM tool fine-tuning (SFT) to M3 | M3 13B · PI-LLM | accuracy: 22.5 -> 71.8 | Improved |
| add ACM tool fine-tuning (SFT) to M3 | M3 13B · NeedleBench-M-RS | accuracy: 30.0 -> 67.6 | Improved |
| add ACM tool fine-tuning (SFT) to M3 | M3 13B · FRAMES | accuracy: 65.2 -> 51.2 | Hurt |
| add dynamic context-aware GSPO RL after SFT (ablation) | Sculptor-M3 · PI-LLM | accuracy: 71.8 -> 99.4 | Improved |
| add dynamic context-aware GSPO RL after SFT (ablation) | Sculptor-M3 · LongBenchV2 | accuracy: 29.2 -> 34.5 | Improved |
| use Sculptor-M3-RL instead of MemAgent | M3 13B · PI-LLM | accuracy: 41.5 -> 99.4 | Improved |
Closest papers by meaning
- Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session Agents (ICLR 2026 · similarity 0.89)
- Can large language models explore in-context? (NeurIPS 2024 · similarity 0.88)
- MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agents (ICLR 2026 · similarity 0.88)
- Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions (ICLR 2026 · similarity 0.88)
- Reasoning Under 1 Billion: Memory-Augmented Reinforcement Learning for Large Language Models (NeurIPS 2026 · similarity 0.87)
- WorldMemArena: Evaluating Multimodal Agent Memory Through Action–World Interaction (NeurIPS 2026 · 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.
