Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
80 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Small Language Model Can Self-correct (2401.07301v2)

Published 14 Jan 2024 in cs.CL and cs.AI

Abstract: Generative LLMs (LMs) such as ChatGPT have exhibited remarkable performance across various downstream tasks. Nevertheless, one of their most prominent drawbacks is generating inaccurate or false information with a confident tone. Previous studies have devised sophisticated pipelines and prompts to induce large LMs to exhibit the capability for self-correction. However, large LMs are explicitly prompted to verify and modify its answers separately rather than completing all steps spontaneously like humans. Moreover, these complex prompts are extremely challenging for small LMs to follow. In this paper, we introduce the \underline{I}ntrinsic \underline{S}elf-\underline{C}orrection (ISC) in generative LLMs, aiming to correct the initial output of LMs in a self-triggered manner, even for those small LMs with 6 billion parameters. Specifically, we devise a pipeline for constructing self-correction data and propose Partial Answer Masking (PAM), aiming to endow the model with the capability for intrinsic self-correction through fine-tuning. We conduct experiments using LMs with parameters sizes ranging from 6 billion to 13 billion in two tasks, including commonsense reasoning and factual knowledge reasoning. Our experiments demonstrate that the outputs generated using ISC outperform those generated without self-correction. We believe that the output quality of even small LMs can be further improved by empowering them with the ability to intrinsic self-correct.

PDF HTML Abstract
Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (33)
  1. S. Welleck, X. Lu, P. West, F. Brahman, T. Shen, D. Khashabi, and Y. Choi, “Generating sequences by learning to self-correct,” ArXiv, vol. abs/2211.00053, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:253244506
  2. J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V. Le, D. Zhou et al., “Chain-of-thought prompting elicits reasoning in large language models,” Advances in Neural Information Processing Systems, vol. 35, pp. 24 824–24 837, 2022.
  3. Y. Weng, M. Zhu, S. He, K. Liu, and J. Zhao, “Large language models are reasoners with self-verification,” arXiv preprint arXiv:2212.09561, 2022.
  4. B. Guo, X. Zhang, Z. Wang, M. Jiang, J. Nie, Y. Ding, J. Yue, and Y. Wu, “How close is chatgpt to human experts? comparison corpus, evaluation, and detection,” 2023.
  5. M. Suzgun, N. Scales, N. Schärli, S. Gehrmann, Y. Tay, H. W. Chung, A. Chowdhery, Q. Le, E. Chi, D. Zhou, and J. Wei, “Challenging BIG-bench tasks and whether chain-of-thought can solve them,” in Findings of the Association for Computational Linguistics: ACL 2023.   Toronto, Canada: Association for Computational Linguistics, Jul. 2023, pp. 13 003–13 051. [Online]. Available: https://aclanthology.org/2023.findings-acl.824
  6. S. Lin, J. Hilton, and O. Evans, “TruthfulQA: Measuring how models mimic human falsehoods,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).   Dublin, Ireland: Association for Computational Linguistics, May 2022, pp. 3214–3252. [Online]. Available: https://aclanthology.org/2022.acl-long.229
  7. T. Liu, Y. Zhang, C. Brockett, Y. Mao, Z. Sui, W. Chen, and B. Dolan, “A token-level reference-free hallucination detection benchmark for free-form text generation,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).   Dublin, Ireland: Association for Computational Linguistics, May 2022, pp. 6723–6737. [Online]. Available: https://aclanthology.org/2022.acl-long.464
  8. P. Manakul, A. Liusie, and M. J. F. Gales, “Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models,” 2023.
  9. Y. Huang, X. Feng, X. Feng, and B. Qin, “The factual inconsistency problem in abstractive text summarization: A survey,” 2023.
  10. J. Wei, M. Bosma, V. Y. Zhao, K. Guu, A. W. Yu, B. Lester, N. Du, A. M. Dai, and Q. V. Le, “Finetuned language models are zero-shot learners,” 2022.
  11. L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. L. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, J. Schulman, J. Hilton, F. Kelton, L. Miller, M. Simens, A. Askell, P. Welinder, P. Christiano, J. Leike, and R. Lowe, “Training language models to follow instructions with human feedback,” 2022.
  12. S. Y. Feng, V. Gangal, J. Wei, S. Chandar, S. Vosoughi, T. Mitamura, and E. Hovy, “A survey of data augmentation approaches for NLP,” in Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021.   Online: Association for Computational Linguistics, Aug. 2021, pp. 968–988. [Online]. Available: https://aclanthology.org/2021.findings-acl.84
  13. D. Zhou, N. Schärli, L. Hou, J. Wei, N. Scales, X. Wang, D. Schuurmans, C. Cui, O. Bousquet, Q. Le, and E. Chi, “Least-to-most prompting enables complex reasoning in large language models,” 2023.
  14. A. Azaria and T. Mitchell, “The internal state of an llm knows when its lying,” 2023.
  15. OpenAI, “Gpt-4 technical report,” 2023.
  16. T. Schick, J. Dwivedi-Yu, Z. Jiang, F. Petroni, P. Lewis, G. Izacard, Q. You, C. Nalmpantis, E. Grave, and S. Riedel, “Peer: A collaborative language model,” in In Proceedings of The 11th International Conference on Learning Representations, 2023.
  17. L. Flower and J. R. Hayes, “A cognitive process theory of writing,” College composition and communication, vol. 32, no. 4, pp. 365–387, 1981.
  18. K. Yang, Y. Tian, N. Peng, and D. Klein, “Re3: Generating longer stories with recursive reprompting and revision,” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing.   Abu Dhabi, United Arab Emirates: Association for Computational Linguistics, Dec. 2022, pp. 4393–4479. [Online]. Available: https://aclanthology.org/2022.emnlp-main.296
  19. D. Paul, M. Ismayilzada, M. Peyrard, B. Borges, A. Bosselut, R. West, and B. Faltings, “Refiner: Reasoning feedback on intermediate representations,” 2023.
  20. A. Madaan, N. Tandon, P. Gupta, S. Hallinan, L. Gao, S. Wiegreffe, U. Alon, N. Dziri, S. Prabhumoye, Y. Yang, S. Gupta, B. P. Majumder, K. Hermann, S. Welleck, A. Yazdanbakhsh, and P. Clark, “Self-refine: Iterative refinement with self-feedback,” 2023.
  21. N. Shinn, F. Cassano, B. Labash, A. Gopinath, K. Narasimhan, and S. Yao, “Reflexion: Language agents with verbal reinforcement learning,” 2023.
  22. Z. Gu, X. Zhu, H. Ye, L. Zhang, J. Wang, S. Jiang, Z. Xiong, Z. Li, Q. He, R. Xu, W. Huang, Z. Wang, S. Wang, W. Zheng, H. Feng, and Y. Xiao, “Xiezhi: An ever-updating benchmark for holistic domain knowledge evaluation,” 2023.
  23. N. Stiennon, L. Ouyang, J. Wu, D. Ziegler, R. Lowe, C. Voss, A. Radford, D. Amodei, and P. F. Christiano, “Learning to summarize with human feedback,” Advances in Neural Information Processing Systems, vol. 33, pp. 3008–3021, 2020.
  24. L. Gao, J. Schulman, and J. Hilton, “Scaling laws for reward model overoptimization,” in International Conference on Machine Learning.   PMLR, 2023, pp. 10 835–10 866.
  25. E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” in International Conference On Learning Representations, 2021.
  26. X. L. Li and P. Liang, “Prefix-tuning: Optimizing continuous prompts for generation,” arXiv preprint arXiv:2101.00190, 2021.
  27. T. Kojima, S. S. Gu, M. Reid, Y. Matsuo, and Y. Iwasawa, “Large language models are zero-shot reasoners,” Advances in neural information processing systems, vol. 35, pp. 22 199–22 213, 2022.
  28. J. Huang, S. S. Gu, L. Hou, Y. Wu, X. Wang, H. Yu, and J. Han, “Large language models can self-improve,” arXiv preprint arXiv:2210.11610, 2022.
  29. D. Ganguli, A. Askell, N. Schiefer, T. Liao, K. Lukošiūtė, A. Chen, A. Goldie, A. Mirhoseini, C. Olsson, D. Hernandez et al., “The capacity for moral self-correction in large language models,” arXiv preprint arXiv:2302.07459, 2023.
  30. Y. Bai, S. Kadavath, S. Kundu, A. Askell, J. Kernion, A. Jones, A. Chen, A. Goldie, A. Mirhoseini, C. McKinnon et al., “Constitutional ai: Harmlessness from ai feedback,” arXiv preprint arXiv:2212.08073, 2022.
  31. J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, “Scaling laws for neural language models,” arXiv preprint arXiv:2001.08361, 2020.
  32. R. Ren, Y. Wang, Y. Qu, W. X. Zhao, J. Liu, H. Tian, H. Wu, J.-R. Wen, and H. Wang, “Investigating the factual knowledge boundary of large language models with retrieval augmentation,” arXiv preprint arXiv:2307.11019, 2023.
  33. A. Holtzman, J. Buys, L. Du, M. Forbes, and Y. Choi, “The curious case of neural text degeneration,” in International Conference on Learning Representations, 2020.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (5)
  1. Haixia Han (4 papers)
  2. Jiaqing Liang (62 papers)
  3. Jie Shi (32 papers)
  4. Qianyu He (26 papers)
  5. Yanghua Xiao (151 papers)
Citations (9)
View on Semantic Scholar