Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
41 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
41 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

Beyond Chain-of-Thought: A Survey of Chain-of-X Paradigms for LLMs (2404.15676v2)

Published 24 Apr 2024 in cs.CL and cs.AI

Abstract: Chain-of-Thought (CoT) has been a widely adopted prompting method, eliciting impressive reasoning abilities of LLMs. Inspired by the sequential thought structure of CoT, a number of Chain-of-X (CoX) methods have been developed to address various challenges across diverse domains and tasks involving LLMs. In this paper, we provide a comprehensive survey of Chain-of-X methods for LLMs in different contexts. Specifically, we categorize them by taxonomies of nodes, i.e., the X in CoX, and application tasks. We also discuss the findings and implications of existing CoX methods, as well as potential future directions. Our survey aims to serve as a detailed and up-to-date resource for researchers seeking to apply the idea of CoT to broader scenarios.

PDF HTML Abstract

Comprehensive Survey of Chain-of-X Methods for Enhancing LLMs Across Diverse Domains

Introduction to Chain-of-X

The paper explores the application of Chain-of-X (CoX) methodologies which expand upon the well-established Chain-of-Thought (CoT) concept in LLMs. CoX is identified as a generalized form of CoT designed to enhance performance on a broader spectrum of tasks beyond basic reasoning. These include but are not limited to multi-modal interaction, hallucination reduction, and complex decision-making across various domains. The diversity in the 'X' or nodes of CoX allows for task-specific adaptations, leading to significant improvements in the handling and execution of complex tasks by LLMs.

Nodes in Chain-of-X

The survey categorizes the nodes used in CoX methods into several distinct types:

  • Intermediates: These nodes extend the CoT concept by incorporating different types of intermediate steps based on task complexities.
  • Augmentation: Nodes that provide supplementary data or directives to enhance the reasoning or decision-making capabilities of LLMs.
  • Feedback: This involves nodes that introduce iterative refinement through feedback, which may come from various external or internal sources.
  • Models: A novel category where a series of specialized models are linked together, each contributing distinct capabilities or knowledge to solve parts of a larger problem.

Each category is designed to tackle specific demands of tasks involving LLMs. For instance, intermediates might focus on decomposing problems into manageable units, while feedback nodes actively refine outputs to enhance accuracy and reliability.

Applied Tasks

Discussing the application areas of CoX methodologies, the paper organizes them into tasks where these methods have shown substantial utility:

  • Multi-Modal Interaction: Techniques like Chain-of-Information and Chain-of-Modality demonstrate improved interactions across different modes of data (text, image, speech).
  • Factuality and Content Safety: CoX methods such as Chain-of-Verification and Chain-of-NLI play crucial roles in reducing hallucinations and aligning model outputs with factual accuracy.
  • Multi-Step Reasoning: CoX frameworks are particularly effective in complex reasoning scenarios, allowing models to address each step with informed precision.
  • Instruction Following: Tailored chains help guide LLMs through structured task execution, interpreting and following complex instructions with higher accuracy.
  • Agency in LLMs: CoX methodologies enable models to act as agents, planning and executing tasks with considerable autonomy and strategic thinking.
  • Evaluation Tools: Innovative CoX frameworks provide new means to test and evaluate the performance of LLMs in various complex scenarios.

Forward-Looking Insights

The paper speculates on future trends and potential improvements in CoX methodologies. For instance, it suggests exploring causal relationships between intermediate nodes and final outputs to better understand the influence of these nodes on the overall performance. Another proposed area of development involves optimizing the inference costs associated with the sequential processing of CoX chains. Further, the application of knowledge distillation through these methods could aid in training more efficient yet smaller models.

In summation, this survey provides a structured and detailed account of how CoX methods can be implemented and leveraged to significantly extend the utility of LLMs across a wide range of tasks and domains. It not only emphasizes the current achievements and utility of CoX methods but also outlines potential avenues for further enhancement and refinement.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (86)
  1. From sparse to dense: GPT-4 summarization with chain of density prompting. In Proceedings of the 4th New Frontiers in Summarization Workshop, pages 68–74, Singapore. Association for Computational Linguistics.
  2. Can knowledge graphs reduce hallucinations in llms?: A survey. arXiv preprint arXiv:2311.07914.
  3. Jinwoo Ahn and Kyuseung Shin. 2024. Recursive chain-of-feedback prevents performance degradation from redundant prompting. arXiv preprint arXiv:2402.02648v2.
  4. Multimodal automated fact-checking: A survey. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 5430–5448, Singapore. Association for Computational Linguistics.
  5. Chain-of-event prompting for multi-document summarization by large language models. International Journal of Web Information Systems.
  6. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17682–17690.
  7. Topologies of reasoning: Demystifying chains, trees, and graphs of thoughts. arXiv preprint arXiv:2401.14295.
  8. Rishabh Bhardwaj and Soujanya Poria. 2023. Red-teaming large language models using chain of utterances for safety-alignment. arXiv preprint arXiv:2308.09662.
  9. A survey on evaluation of large language models. ACM Transactions on Intelligent Systems and Technology.
  10. A survey of chain of thought reasoning: Advances, frontiers and future. arXiv preprint arXiv:2309.15402.
  11. Chain-of-verification reduces hallucination in large language models. arXiv preprint arXiv:2309.11495.
  12. Choire: Characterizing and predicting human opinions with chain of opinion reasoning. arXiv preprint arXiv:2311.08385.
  13. Attacks, defenses and evaluations for llm conversation safety: A survey. arXiv preprint arXiv:2402.09283.
  14. Efficient tool use with chain-of-abstraction reasoning. arXiv preprint arXiv:2401.17464.
  15. Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997.
  16. Peiyuan Gong and Jiaxin Mao. 2023. Coascore: Chain-of-aspects prompting for nlg evaluation. arXiv preprint arXiv:2312.10355.
  17. Chain-of-interaction: Enhancing large language models for psychiatric behavior understanding by dyadic contexts. arXiv preprint arXiv:2403.13786.
  18. Chain-of-instructions: Compositional instruction tuning on large language models. arXiv preprint arXiv:2402.11532.
  19. Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes. In Findings of the Association for Computational Linguistics: ACL 2023, pages 8003–8017, Toronto, Canada. Association for Computational Linguistics.
  20. Chatdb: Augmenting llms with databases as their symbolic memory. arXiv preprint arXiv:2306.03901.
  21. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  22. Chain-of-symbol prompting elicits planning in large langauge models. arXiv preprint arXiv:2305.10276.
  23. Chain of explanation: New prompting method to generate quality natural language explanation for implicit hate speech. In Companion Proceedings of the ACM Web Conference 2023, pages 90–93.
  24. Jie Huang and Kevin Chen-Chuan Chang. 2023. Towards reasoning in large language models: A survey. In Findings of the Association for Computational Linguistics: ACL 2023, pages 1049–1065, Toronto, Canada. Association for Computational Linguistics.
  25. Coq: An empirical framework for multi-hop question answering empowered by large language models. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 11566–11570. IEEE.
  26. Generalizing visual question answering from synthetic to human-written questions via a chain of qa with a large language model. arXiv preprint arXiv:2401.06400.
  27. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199–22213.
  28. Chain of reference prompting helps llm to think like a lawyer. In Generative AI+ Law Workshop.
  29. Code simulation challenges for large language models. arXiv preprint arXiv:2401.09074.
  30. Codechain: Towards modular code generation through chain of self-revisions with representative sub-modules. In The Twelfth International Conference on Learning Representations.
  31. Rlaif: Scaling reinforcement learning from human feedback with ai feedback. arXiv preprint arXiv:2309.00267.
  32. Chain of empathy: Enhancing empathetic response of large language models based on psychotherapy models. arXiv preprint arXiv:2311.04915.
  33. Chain of natural language inference for reducing large language model ungrounded hallucinations. arXiv preprint arXiv:2310.03951.
  34. Chain of code: Reasoning with a language model-augmented code emulator. arXiv preprint arXiv:2312.04474.
  35. Symbolic chain-of-thought distillation: Small models can also “think” step-by-step. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2665–2679, Toronto, Canada. Association for Computational Linguistics.
  36. Chain-of-knowledge: Grounding large language models via dynamic knowledge adapting over heterogeneous sources. In The Twelfth International Conference on Learning Representations.
  37. Ecomgpt: Instruction-tuning large language models with chain-of-task tasks for e-commerce. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 18582–18590.
  38. Chain of hindsight aligns language models with feedback. In The Twelfth International Conference on Learning Representations.
  39. Chain-of-spot: Interactive reasoning improves large vision-language models. arXiv preprint arXiv:2403.12966.
  40. Chain-of-dictionary prompting elicits translation in large language models. arXiv preprint arXiv:2305.06575.
  41. Chain of history: Learning and forecasting with llms for temporal knowledge graph completion. arXiv preprint arXiv:2401.06072.
  42. Faithful chain-of-thought reasoning. In Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 305–329, Nusa Dua, Bali. Association for Computational Linguistics.
  43. Large language models play starcraft ii: Benchmarks and a chain of summarization approach. arXiv preprint arXiv:2312.11865.
  44. Chain of images for intuitively reasoning. arXiv preprint arXiv:2311.09241.
  45. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:27730–27744.
  46. Chain-of-action: Faithful and multimodal question answering through large language models. arXiv preprint arXiv:2403.17359.
  47. Cogcom: Train large vision-language models diving into details through chain of manipulations. arXiv preprint arXiv:2402.04236.
  48. Chain-of-lora: Enhancing the instruction fine-tuning performance of low-rank adaptation on diverse instruction set. IEEE Signal Processing Letters.
  49. Scaling language models: Methods, analysis & insights from training gopher. arXiv preprint arXiv:2112.11446.
  50. Chain of logic: Rule-based reasoning with large language models. arXiv preprint arXiv:2402.10400.
  51. Chain-of-discussion: A multi-model framework for complex evidence-based question answering. arXiv preprint arXiv:2402.16313.
  52. Advancing large multi-modal models with explicit chain-of-reasoning and visual question generation. arXiv preprint arXiv:2401.10005.
  53. Towards understanding chain-of-thought prompting: An empirical study of what matters. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2717–2739, Toronto, Canada. Association for Computational Linguistics.
  54. Intervenor: Prompt the coding ability of large language models with the interactive chain of repairing. arXiv preprint arXiv:2311.09868.
  55. Boosting language models reasoning with chain-of-knowledge prompting. arXiv preprint arXiv:2306.06427.
  56. Self-consistency improves chain of thought reasoning in language models. In The Eleventh International Conference on Learning Representations.
  57. Aligning large language models with human: A survey. arXiv preprint arXiv:2307.12966.
  58. Chain-of-table: Evolving tables in the reasoning chain for table understanding. In The Twelfth International Conference on Learning Representations.
  59. Chain of thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems.
  60. Chain-of-look prompting for verb-centric surgical triplet recognition in endoscopic videos. In Proceedings of the 31st ACM International Conference on Multimedia, pages 5007–5016.
  61. The rise and potential of large language model based agents: A survey. arXiv preprint arXiv:2309.07864.
  62. Chain of lora: Efficient fine-tuning of language models via residual learning. arXiv preprint arXiv:2401.04151.
  63. Which llm to play? convergence-aware online model selection with time-increasing bandits. arXiv preprint arXiv:2403.07213.
  64. Hallucination diversity-aware active learning for text summarization. arXiv preprint arXiv:2404.01588.
  65. Enhancing temporal knowledge graph forecasting with large language models via chain-of-history reasoning. arXiv preprint arXiv:2402.14382.
  66. Unified human-scene interaction via prompted chain-of-contacts. In The Twelfth International Conference on Learning Representations.
  67. Chain-of-experts: When LLMs meet complex operations research problems. In The Twelfth International Conference on Learning Representations.
  68. Search-in-the-chain: Towards the accurate, credible and traceable content generation for complex knowledge-intensive tasks. arXiv preprint arXiv:2304.14732.
  69. L3go: Language agents with chain-of-3d-thoughts for generating unconventional objects. arXiv preprint arXiv:2402.09052.
  70. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36.
  71. Chain-of-note: Enhancing robustness in retrieval-augmented language models. arXiv preprint arXiv:2311.09210.
  72. Towards better chain-of-thought prompting strategies: A survey. arXiv preprint arXiv:2310.04959.
  73. Large language models meet NL2Code: A survey. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7443–7464, Toronto, Canada. Association for Computational Linguistics.
  74. Chain-of-layer: Iteratively prompting large language models for taxonomy induction from limited examples. arXiv preprint arXiv:2402.07386.
  75. Tablegpt: Towards unifying tables, nature language and commands into one gpt. arXiv preprint arXiv:2307.08674.
  76. Zhuosheng Zhan and Aston Zhang. 2023. You only look at screens: Multimodal chain-of-action agents. arXiv preprint arXiv:2309.11436.
  77. SpeechGPT: Empowering large language models with intrinsic cross-modal conversational abilities. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 15757–15773, Singapore. Association for Computational Linguistics.
  78. Speechgpt-gen: Scaling chain-of-information speech generation. arXiv preprint arXiv:2401.13527.
  79. Vision-language models for vision tasks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence.
  80. Android in the zoo: Chain-of-action-thought for gui agents. arXiv preprint arXiv:2403.02713.
  81. Instruction tuning for large language models: A survey. arXiv preprint arXiv:2308.10792.
  82. Siren’s song in the ai ocean: a survey on hallucination in large language models. arXiv preprint arXiv:2309.01219.
  83. Coie: Chain-of-instruct editing for multi-attribute face manipulation. arXiv preprint arXiv:2312.07879.
  84. Retrieving multimodal information for augmented generation: A survey. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4736–4756, Singapore. Association for Computational Linguistics.
  85. Least-to-most prompting enables complex reasoning in large language models. In The Eleventh International Conference on Learning Representations.
  86. Minedreamer: Learning to follow instructions via chain-of-imagination for simulated-world control. arXiv preprint arXiv:2403.12037.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Yu Xia (65 papers)
  2. Rui Wang (996 papers)
  3. Xu Liu (213 papers)
  4. Mingyan Li (12 papers)
  5. Tong Yu (119 papers)
  6. Xiang Chen (343 papers)
  7. Julian McAuley (238 papers)
  8. Shuai Li (295 papers)
Citations (9)
View on Semantic Scholar