Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
Internal consistency and self-feedback in large language models: A survey
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
abstract
Large language models (LLMs) often exhibit deficient reasoning or generate hallucinations. To address these, studies prefixed with "Self-" such as Self-Consistency, Self-Improve, and Self-Refine have been initiated. They share a commonality: involving LLMs evaluating and updating themselves. Nonetheless, these efforts lack a unified perspective on summarization, as existing surveys predominantly focus on categorization. In this paper, we use a unified perspective of internal consistency, offering explanations for reasoning deficiencies and hallucinations. Internal consistency refers to the consistency in expressions among LLMs' latent, decoding, or response layers based on sampling methodologies. Then, we introduce an effective theoretical framework capable of mining internal consistency, named Self-Feedback. This framework consists of two modules: Self-Evaluation and Self-Update. The former captures internal consistency signals, while the latter leverages the signals to enhance either the model's response or the model itself. This framework has been employed in numerous studies. We systematically classify these studies by tasks and lines of work; summarize relevant evaluation methods and benchmarks; and delve into the concern, "Does Self-Feedback Really Work?" We also propose several critical viewpoints, including the "Hourglass Evolution of Internal Consistency", "Consistency Is (Almost) Correctness" hypothesis, and "The Paradox of Latent and Explicit Reasoning". The relevant resources are open-sourced at https://github.com/IAAR-Shanghai/ICSFSurvey.
representative citing papers
Deliberative Searcher integrates retrieval search, multi-step verification, and RL training with a soft reliability constraint to improve alignment between LLM confidence and correctness in open-domain QA.
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
Introduces Wasserstein equilibrium decoding that improves accuracy and convergence speed for small VLMs on medical VQA benchmarks by using semantic consensus instead of lexical order.
LLMs settle on their answer after a minority of CoT tokens and produce an average 760 more as post-decision explanation, enabling early stopping that saves 500 tokens per query at a 2% accuracy cost.
FVRuleLearner retrieves learned operator-level reasoning rules to boost the functional correctness of LLM-generated SystemVerilog assertions by roughly 30 percentage points over simple prompting baselines.
Temperature and persona variations shape consensus speed in LLM multi-agent coding but produce no robust accuracy gains over single agents on human-annotated tutoring transcripts.
LoRA continued pretraining on a small U.S. transportation corpus lifts BLEU-4 and ROUGE for Qwen2.5-7B and LLaMA-3.1-8B far above the other four models tested.
citing papers explorer
-
Reasoning Portability: Guiding Continual Learning for MLLMs in the RLVR Era
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
-
Deliberative Searcher: Improving LLM Reliability via Reinforcement Learning with constraints
Deliberative Searcher integrates retrieval search, multi-step verification, and RL training with a soft reliability constraint to improve alignment between LLM confidence and correctness in open-domain QA.
-
Unified Audio Intelligence Without Regressing on Text Intelligence
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
-
Wasserstein Equilibrium Decoding for Reliable Medical Visual Question Answering
Introduces Wasserstein equilibrium decoding that improves accuracy and convergence speed for small VLMs on medical VQA benchmarks by using semantic consensus instead of lexical order.
-
Large Language Models Decide Early and Explain Later
LLMs settle on their answer after a minority of CoT tokens and produce an average 760 more as post-decision explanation, enabling early stopping that saves 500 tokens per query at a 2% accuracy cost.
-
FVRuleLearner: Operator-Level Reasoning Tree (Op-Tree)-Based Rules Learning for Formal Verification
FVRuleLearner retrieves learned operator-level reasoning rules to boost the functional correctness of LLM-generated SystemVerilog assertions by roughly 30 percentage points over simple prompting baselines.
-
Temperature and Persona Shape LLM Agent Consensus With Minimal Accuracy Gains in Qualitative Coding
Temperature and persona variations shape consensus speed in LLM multi-agent coding but produce no robust accuracy gains over single agents on human-annotated tutoring transcripts.
-
Customized Generative AI Agent for Transportation Engineering Practice: A Development and Continued Pre-training Guideline
LoRA continued pretraining on a small U.S. transportation corpus lifts BLEU-4 and ROUGE for Qwen2.5-7B and LLaMA-3.1-8B far above the other four models tested.