mlsanityData to 14 Oct 2026

SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses

arXiv 2404.04298 ↗First posted 4 Apr 2024cs.AI cs.CL cs.LGcited by 26

On the map: Synthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training

Abstract (arXiv)

Can LLMs consistently improve their previous outputs for better results? For this to be true, LLMs would need to be better at discriminating among previously-generated alternatives, than generating initial responses. We explore the validity of this hypothesis in practice. We first formulate a unified framework that allows us to compare the generative and discriminative capability of any model on any task. In our resulting experimental analysis of several open-source and industrial LLMs, we observe that models are not reliably better at discriminating among previously-generated alternatives than generating initial responses. This finding challenges the notion that LLMs may be able to enhance their performance only through their own judgment.

Key facts. Cited by 26 papers in the map (data through 14 Oct 2026). Placed in Self-rewarding alignment and self-training, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 41 of its 53 references that are in the map. First posted on arXiv 4 Apr 2024.

Builds on

41 of 53 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay7 Feb
  2. 2024Self-Rewarding Language Models18 Jan
  3. 2024Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models2 Jan

Other references · 38

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  5. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  6. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  7. 2022Scaling Instruction-Finetuned Language Modelscited by 3,024
  8. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  9. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  10. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,724
  11. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  12. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
26 more
  1. 2023ChatGPT Outperforms Crowd-Workers for Text-Annotation Taskscited by 870
  2. 2022Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Taskscited by 868
  3. 2023CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquingcited by 783
  4. 2023The Flan Collection: Designing Data and Methods for Effective Instruction Tuningcited by 739
  5. 2023Can Large Language Models Be an Alternative to Human Evaluations?cited by 734
  6. 2022Self-critiquing models for assisting human evaluatorscited by 380
  7. 2016SeqGAN: Sequence Generative Adversarial Nets with Policy Gradientcited by 376
  8. 2022PromptSource: An Integrated Development Environment and Repository for Natural Language Promptscited by 289
  9. 2022Generating Sequences by Learning to Self-Correctcited by 277
  10. 2023Automatically Correcting Large Language Models: Surveying the landscape of diverse self-correction strategiescited by 238
  11. 2023Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solvecited by 178
  12. 2024On the Self-Verification Limitations of Large Language Models on Reasoning and Planning Taskscited by 142
  13. 2024When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboardscited by 137
  14. 2023Can Large Language Models Really Improve by Self-critiquing Their Own Plans?cited by 108
  15. 2023The Generative AI Paradox: "What It Can Create, It May Not Understand"cited by 98
  16. 2023GPT-4 Doesn't Know It's Wrong: An Analysis of Iterative Prompting for Reasoning Problemscited by 89
  17. 2024CriticBench: Benchmarking LLMs for Critique-Correct Reasoningcited by 88
  18. 2023LLMs cannot find reasoning errors, but can correct them given the error locationcited by 85
  19. 2023Benchmarking Generation and Evaluation Capabilities of Large Language Models for Instruction Controllable Summarizationcited by 69
  20. 2022On the Blind Spots of Model-Based Evaluation Metrics for Text Generationcited by 49
  21. 2024When is Tree Search Useful for LLM Planning? It Depends on the Discriminatorcited by 47
  22. 2023Gaining Wisdom from Setbacks: Aligning Large Language Models via Mistake Analysiscited by 35
  23. 2017Adversarial Generation of Natural Languagecited by 33
  24. 2023Crystal: Introspective Reasoners Reinforced with Self-Feedbackcited by 15
  25. 2023On the Intersection of Self-Correction and Trust in Language Modelscited by 12
  26. 2023Learning and Leveraging Verifiers to Improve Planning Capabilities of Pre-trained Language Modelscited by 10

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 6; 2025: 8; 2026: 12.

26 papers.

Sort:

By area:

  1. 2024ScienceAgentBench: Toward Rigorous Assessment of Language Agents for Data-Driven Scientific Discoverycited by 235
  2. 2025ScoreFlow: Mastering LLM Agent Workflows via Score-based Preference Optimizationcited by 56
  3. 2026SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learningcited by 32
  4. 2024Large Language Models Can Self-Improve in Long-context Reasoningcited by 29
  5. 2024Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Modelscited by 18
  6. 2026DeepPresenter: Environment-Grounded Reflection for Agentic Presentation Generationcited by 18
  7. 2025Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMscited by 14
  8. 2026Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skillscited by 10
  9. 2024RATIONALYST: Mining Implicit Rationales for Process Supervision of Reasoningcited by 6
  10. 2024Beyond Introspection: Reinforcing Thinking via Externalist Behavioral Feedbackcited by 5
  11. 2025RankAlign: A Ranking View of the Generator-Validator Gap in Large Language Modelscited by 4
  12. 2024Memory-Augmented Agent Training for Business Document Understandingcited by 3
14 more
  1. 2026Judging Is Not Enumerating: Silent Omissions in LLM-Authored Acceptable Setscited by 2
  2. 2025GuessBench: Sensemaking Multimodal Creativity in the Wildcited by 1
  3. 2026Who Flips? Self- and Cross-Model Counterarguments Reveal Answer Instability in LLMscited by 1
  4. 2026Can LLMs Judge Better Than They Generate? Evaluating Task Asymmetry, Mechanistic Interpretability and Transferability for In-Context QAcited by 1
  5. 2025A Data-Centric Approach for Safe and Secure Large Language Models against Threatening and Toxic Contentcited by 0
  6. 2025Unlocking Recursive Thinking of LLMs: Alignment via Refinementcited by 0
  7. 2025A Rolling Stone Gathers No Moss: Adaptive Policy Optimization for Stable Self-Evaluation in Large Multimodal Modelscited by 0
  8. 2025Safety Game: Inference-Time Alignment of Black-Box LLMs via Constrained Optimizationcited by 0
  9. 2026Aligning to Illusions: Choice Blindness in Human and AI Feedbackcited by 0
  10. 2026DCGC: Draft-Conditioned Global Correction for Complex Reasoning with Masked Diffusion Modelscited by 0
  11. 2026Reflective Recovery: A Self-Supervised Method for Reasoning by Learning from Mistakescited by 0
  12. 2026The Architect, the Adversary, and the Judge: Closed-Loop Generation of Standards-Aligned Assessment Items at Scalecited by 0
  13. 2026Multi-agent discussion gains less when dissent is withheldcited by 0
  14. 2026Feedback as an Interface: Controlling Form and Detail in LLM Translation Refinementcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Direct-Inverse Prompting: Analyzing LLMs' Discriminative Capacity in Self-Improving Generation
  2. 2024LLM Evaluators Recognize and Favor Their Own Generations
  3. 2025Do LLM Evaluators Prefer Themselves for a Reason?
  4. 2025LLMs Don't Know Their Own Decision Boundaries: The Unreliability of Self-Generated Counterfactual Explanations
  5. 2025LLMs Can Generate a Better Answer by Aggregating Their Own Responses
  6. 2024Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement
  7. 2025The Self-Execution Benchmark: Measuring LLMs' Attempts to Overcome Their Lack of Self-Execution
  8. 2026Evaluating LLMs as Human Surrogates in Controlled Experiments