Research map / Large language models / Agents, code and math
Codified Finite-state Machines for Role-playing
Letian Peng, Yupeng Hou, Kun Zhou, Jingbo Shang
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 codified FSM state tracking | Codified Profile · Fandom Benchmark main | NLI score: 81.36 -> 82.65 | Improved |
| summarization-based state update (PromptTrans) | Codified Profile · Fandom Benchmark main | NLI score: 81.36 -> 81.23 | No clear effect |
| remove state registration (ablation) | Codified FSM · Fandom Benchmark merged | NLI score: 82.58 -> 83.67 | Improved |
| keep Other and Unactivated fallback states (ablation) | Codified FSM · Fandom Benchmark merged | NLI score: 82.92 / 83.17 -> 83.67 | Improved |
| include personality FSMs (ablation) | Codified FSM · Fandom Benchmark merged | NLI score: 82.62 -> 83.67 | Improved |
| use probabilistic transitions (CPFSM) | Codified FSM · Fandom Benchmark merged, llama-3.2-1b-it | NLI score: 62.77 -> 63.32 | Improved |
| coarser transition step (3 sentences per transition) (ablation) | Codified FSM · Fandom Benchmark merged | NLI score: 83.67 -> 82.95 | Hurt |
| use O(n+k) codification instead of O(n^2) (ablation) | CPFSM · Fandom Benchmark merged, 1B | NLI score: 64.14 -> 63.43; questions 50.00 -> 17.71 | No clear effect |
| execute codified FSM instead of PromptTrans | PromptTrans · Fandom Benchmark merged | forward passes per turn: 130.78 -> 30.17 | Improved |
| add dynamic state maintenance | Codified FSM · Fandom Benchmark main / minor | NLI score: 82.65 -> 83.04 (main); 84.60 -> 84.48 (minor) | Mixed |
| use open-source qwen3-coder as codifier (ablation) | CFSM · Fandom Benchmark main | NLI score: 62.04 -> 60.56 | Hurt |
| add chain-of-thought to prompting | gpt-4.1 state tracking · Synthetic FSM paths | accuracy | Mixed |
Closest papers by meaning
- Codifying Character Logic in Role-Playing (NeurIPS 2025 · similarity 0.90)
- CoSER: Coordinating LLM-Based Persona Simulation of Established Roles (ICML 2025 · similarity 0.87)
- Do LLMs Build World Representations? Probing Through the Lens of State Abstraction (NeurIPS 2024 · similarity 0.86)
- Concept Incongruence: An Exploration of Time and Death in Role Playing (NeurIPS 2025 · similarity 0.86)
- Quantifying and Optimizing Global Faithfulness in Persona-driven Role-playing (NeurIPS 2024 · similarity 0.85)
- DMT-RoleBench: A Dynamic Multi-Turn Dialogue Based Benchmark for Role-Playing Evaluation of Large Language Model and Agent (AAAI 2025 · similarity 0.84)
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.
