AI Paper Atlas

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

Intuitive Thinking: Expanding Large Language Models’ Thinking for Rapid Decision-Making on Candidate Corrections in Chinese Grammar Error Correction

Lintao Long, Ruizhang Huang, Ruina Bai, Yongbin Qin, Qihang Fu

AAAI 2026 · Large language models › Agents, code and math · approach: Agent instruction-following benchmarks · 0 citations (OpenAlex)

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
inject LLM token confidence into correction embeddings of the lightweight ExIT model (ablation)ExIT (Qwen2.5-14B) · FCGEC-testF0.5: 50.65 -> 58.63Improved
add LLM alignment loss on top of confidence injection (ablation)ExIT (Qwen2.5-14B) · FCGEC-testF0.5: 58.63 -> 59.12Improved
add LLM alignment loss alone, without confidence injection (ablation)ExIT (Qwen2.5-14B) · FCGEC-testF0.5: 50.65 -> 50.67No clear effect
apply ExIT reranking over sampled candidatesQwen2.5-14B fine-tuned LLM · FCGEC-testF0.5: 52.98 -> 59.12Improved
set LLM sampling temperature to 0 (greedy)Qwen2.5-14B fine-tuned LLM · FCGEC-testF0.5: 52.98 -> 56.05Improved
ExIT instead of greedy decoding at temperature 0Qwen2.5-14B fine-tuned LLM · FCGEC-testF0.5: 56.05 -> 59.12Improved
train 14B with ExIT on small data instead of 7B on large-scale dataLLM CGEC training · FCGEC-testF0.5Improved
apply ExIT trained on Qwen2.5-14B to other LLMsExIT · FCGEC-test, other LLMsF0.5Improved
use ExIT over Qwen2-7B-EPO on FCGEC-testQwen2-7B · FCGEC-testF0.5: 59.63 (EPO) vs 58.50 (ExIT)No clear effect
scale base LLM from 7B to 14B in ExITExIT (Qwen2.5) · FCGEC-testF0.5: 56.17 -> 59.12Improved

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