AI Paper Atlas

Research map / Large language models / Agents, code and math

Codified Finite-state Machines for Role-playing

Letian Peng, Yupeng Hou, Kun Zhou, Jingbo Shang

ICLR 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents

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 changedWhereMeasuredEffect
add codified FSM state trackingCodified Profile · Fandom Benchmark mainNLI score: 81.36 -> 82.65Improved
summarization-based state update (PromptTrans)Codified Profile · Fandom Benchmark mainNLI score: 81.36 -> 81.23No clear effect
remove state registration (ablation)Codified FSM · Fandom Benchmark mergedNLI score: 82.58 -> 83.67Improved
keep Other and Unactivated fallback states (ablation)Codified FSM · Fandom Benchmark mergedNLI score: 82.92 / 83.17 -> 83.67Improved
include personality FSMs (ablation)Codified FSM · Fandom Benchmark mergedNLI score: 82.62 -> 83.67Improved
use probabilistic transitions (CPFSM)Codified FSM · Fandom Benchmark merged, llama-3.2-1b-itNLI score: 62.77 -> 63.32Improved
coarser transition step (3 sentences per transition) (ablation)Codified FSM · Fandom Benchmark mergedNLI score: 83.67 -> 82.95Hurt
use O(n+k) codification instead of O(n^2) (ablation)CPFSM · Fandom Benchmark merged, 1BNLI score: 64.14 -> 63.43; questions 50.00 -> 17.71No clear effect
execute codified FSM instead of PromptTransPromptTrans · Fandom Benchmark mergedforward passes per turn: 130.78 -> 30.17Improved
add dynamic state maintenanceCodified FSM · Fandom Benchmark main / minorNLI score: 82.65 -> 83.04 (main); 84.60 -> 84.48 (minor)Mixed
use open-source qwen3-coder as codifier (ablation)CFSM · Fandom Benchmark mainNLI score: 62.04 -> 60.56Hurt
add chain-of-thought to promptinggpt-4.1 state tracking · Synthetic FSM pathsaccuracyMixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy