Elamur Research⌕
☰
home / papers / 2024 / The pitfalls of next-token prediction

The pitfalls of next-token prediction

Paper · 2024 · Gregor Bachmann and Vaishnavh Nagarajan

Summary

The authors test whether a model trained only to predict the next token can solve tasks where the answer has to be planned in advance. They propose distinguishing two regimes: autoregressive generation at inference time and training with teacher-forcing. The common criticism about error accumulation applies only to the first regime. It tacitly assumes that training has already produced an accurate next-token predictor. The authors show that in "lookahead" tasks this assumption does not hold. Teacher-forcing lets the model guess later tokens from the already revealed part of the correct answer (Clever Hans cheat). As a result, the model can barely learn the early tokens, which are exactly where planning is needed. To test this, the authors built a minimal path-finding task on a graph that is easy to solve. Yet both the Transformer and Mamba fail on it, even on examples from the same distribution as the training data. Preliminary experiments show that teacherless training eliminates this flaw in some settings. In this approach, the model predicts several future tokens at once and receives dummy tokens as input. The paper provides a clear basis for the discussion of the limits of next-token prediction and for the search for alternative training methods.

Key points

  • The classic example of error accumulation: if the per-token error is 0.01, the probability of at least one error over 200 tokens reaches about 0.86. The authors stress that the failure they describe is different: the model errs right at the start of the answer, not gradually.
  • Two concepts are proposed to describe the training failure: Clever Hans cheat (the model "peeks" at the revealed prefix of the correct answer) and Indecipherable Token failure (early tokens, lacking this hint, become nearly unlearnable).
  • The failure is shown empirically for two architectures: Transformer (Vaswani et al., 2017) and Mamba (Gu & Dao, 2023). Hence the problem lies in the training objective itself, not in the architecture or the generation method. A reversed answer sequence and chain-of-thought-style supervision also help.
  • Teacherless training is borrowed from earlier works where it was used for a different purpose: Monea et al., 2023 ("parallel speculative sampling" for faster inference) and Tschannen et al., 2023. A close analogue from imitation learning is action chunking (Zhao et al., 2023).
  • The paper appeared at ICML 2024 (PMLR 235); the authors are from ETH Zürich and Google Research. The code is openly available in the GitHub repository gregorbachmann/Next-Token-Failures.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢GoogleMountain View · United States533 works · 443,885 citations🎓ETH ZurichZurich · Switzerland97 works · 9,612 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 works

Cite

Cite as Gregor Bachmann and Vaishnavh Nagarajan. 2024. The pitfalls of next-token prediction. arXiv:2403.06963.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2403.06963 · arXiv 2403.06963

@misc{bachmann2024pitfalls,
  title = {{The pitfalls of next-token prediction}},
  author = {Gregor Bachmann and Vaishnavh Nagarajan},
  year = {2024},
  doi = {10.48550/arxiv.2403.06963},
  eprint = {2403.06963},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2403.06963}
}

References

108 references, 32 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alabdulmohsin, I., Tran, V. Q., and Dehghani, M. Fractal patterns may unravel the intelligence in next-token prediction, 2024. find
  2. Allen-Zhu, Z. and Li, Y. Physics of language models: Part 3.2, knowledge manipulation. arXiv preprint arXiv:2309.14402, 2023.
  3. Anil, C., Wu, Y., Andreassen, A., Lewkowycz, A., Misra, V., Ramasesh, V. V., Slone, A., Gur-Ari, G., Dyer, E., and Neyshabur, B. Exploring length generalization in large language models. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Syste…
  4. Arkoudas, K. Chatgpt is no stochastic parrot. but it also claims that 1 is greater than 1. Philosophy & Technology, 36(3):54, 2023. find
  5. Artetxe, M., Du, J., Goyal, N., Zettlemoyer, L., and Stoyanov, V. On the role of bidirectionality in language model pre-training. In Findings of the Association for Computational Linguistics: EMNLP 2022, Abu Dhabi, United Arab Emirates, December 7-11, 2022, pp. 3973–3985. Association for Computation… find
  6. Bahdanau, D., Brakel, P., Xu, K., Goyal, A., Lowe, R., Pineau, J., Courville, A. C., and Bengio, Y. An actor-critic algorithm for sequence prediction. In 5th International Conference on Learning Representations, ICLR 2017, Conference Track Proceedings, 2017.
  7. Bender, E. M., Gebru, T., McMillan-Major, A., and Shmitchell, S. On the dangers of stochastic parrots: Can language models be too big? In FAccT ’21: 2021 ACM Conference on Fairness, Accountability, and Transparency, Virtual Event / Toronto, Canada, March 3-10, 2021, pp. 610–623. find
  8. Bengio, S., Vinyals, O., Jaitly, N., and Shazeer, N. Scheduled sampling for sequence prediction with recurrent neural networks. In Advances in Neural Information Processing Systems 28: Annual Conference on Neural Information Processing Systems 2015, pp. 1171–1179, 2015. find
  9. Besta, M., Blach, N., Kubicek, A., Gerstenberger, R., Podstawski, M., Gianinazzi, L., Gajda, J., Lehmann, T., Niewiadomski, H., Nyczyk, P., et al. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 20… find
  10. Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., et al. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712, 2023.
  11. Burtsev, M. S., Kuratov, Y., Peganov, A., and Sapunov, G. V. Memory transformer. arXiv preprint arXiv:2006.11527, 2020. — [arXiv:2006.11527](https://arxiv.org/abs/2006.11527) find
  12. Chang, K., Krishnamurthy, A., Agarwal, A., III, H. D., and Langford, J. Learning to search better than your teacher. In Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, volume 37 of JMLR Workshop and Conference Proceedings, 2015. find
  13. Cobbe, K., Kosaraju, V., Bavarian, M., Chen, M., Jun, H., Kaiser, L., Plappert, M., Tworek, J., Hilton, J., Nakano, R., Hesse, C., and Schulman, J. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021.
  14. Daumé III, H., Langford, J., and Marcu, D. Search-based structured prediction. Mach. Learn., 75(3):297–325, 2009. find
  15. Dawid, A. and LeCun, Y. Introduction to latent variable energy-based models: A path towards autonomous machine intelligence. arXiv preprint arXiv:2306.02572, 2023. find
  16. Du, L., Mei, H., and Eisner, J. Autoregressive modeling with lookahead attention. arXiv preprint arXiv:2305.12272, 2023a. — [arXiv:2305.12272](https://arxiv.org/abs/2305.12272) find
  17. Du, L., Torroba Hennigen, L., Pimentel, T., Meister, C., Eisner, J., and Cotterell, R. A measure-theoretic characterization of tight language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Toronto, Canada, 2023b. Associatio… find
  18. Dziri, N., Lu, X., Sclar, M., Li, X. L., Jiang, L., Lin, B. Y., Welleck, S., West, P., Bhagavatula, C., Le Bras, R., et al. Faith and fate: Limits of transformers on compositionality. Advances in Neural Information Processing Systems, 36, 2024.
  19. Feng, G., Zhang, B., Gu, Y., Ye, H., He, D., and Wang, L. Towards revealing the mystery behind chain of thought: a theoretical perspective. Advances in Neural Information Processing Systems, 36, 2023. find
  20. Glasmachers, T. Limits of end-to-end learning. In Zhang, M. and Noh, Y. (eds.), Proceedings of The 9th Asian Conference on Machine Learning, ACML 2017, volume 77 of Proceedings of Machine Learning Research, pp. 17–32. find
  21. Gokaslan, A. and Cohen, V. Openwebtext corpus. http://Skylion007.github.io/OpenWebTextCorpus, 2019. ↗
  22. Gong, S., Li, M., Feng, J., Wu, Z., and Kong, L. Diffuseq: Sequence to sequence text generation with diffusion models. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023.
  23. Goyal, A., Lamb, A., Zhang, Y., Zhang, S., Courville, A. C., and Bengio, Y. Professor forcing: A new algorithm for training recurrent networks. In Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016, pp. 4601– 4609, 2016. find
  24. Goyal, S., Ji, Z., Rawat, A. S., Menon, A. K., Kumar, S., and Nagarajan, V. Think before you speak: Training language models with pause tokens. The Twelfth International Conference on Learning Representations, ICLR 2024, 20234. find
  25. Gu, A. and Dao, T. Mamba: Linear-time sequence modeling with selective state spaces, 2023. find
  26. Gu, J., Bradbury, J., Xiong, C., Li, V. O. K., and Socher, R. Non-autoregressive neural machine translation. In 6th International Conference on Learning Representations, ICLR 2018, Conference Track Proceedings.
  27. Gülçehre, Ç. and Bengio, Y. Knowledge matters: Importance of prior information for optimization. J. Mach. Learn. Res., 17:8:1–8:32, 2016. find
  28. Gurnee, W., Nanda, N., Pauly, M., Harvey, K., Troitskii, D., and Bertsimas, D. Finding neurons in a haystack: Case studies with sparse probing. arXiv preprint arXiv:2305.01610, 2023.
  29. Havrilla, A., Du, Y., Raparthy, S. C., Nalmpantis, C., Dwivedi-Yu, J., Zhuravinskyi, M., Hambro, E., Sukhbaatar, S., and Raileanu, R. Teaching large language models to reason with reinforcement learning, 2024.
  30. Hsieh, C.-Y., Li, C.-L., Yeh, C.-K., Nakhost, H., Fujii, Y., Ratner, A., Krishna, R., Lee, C.-Y., and Pfister, T. Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes. arXiv preprint arXiv:2305.02301, 2023.
  31. Huang, W., Xia, F., Xiao, T., Chan, H., Liang, J., Florence, P., Zeng, A., Tompson, J., Mordatch, I., Chebotar, Y., Sermanet, P., Jackson, T., Brown, N., Luu, L., Levine, S., Hausman, K., and Ichter, B. Inner monologue: Embodied reasoning through planning with language models. In Conference on Robot… find
  32. Jiang, A. Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D. S., de las Casas, D., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., Lavaud, L. R., Lachaux, M.-A., Stock, P., Scao, T. L., Lavril, T., Wang, T., Lacroix, T., and Sayed, W. E. Mistral 7b, 2023. find
  33. Kääriäinen, M. Lower bounds for reductions. In Atomic Learning Workshop, 2006. find
  34. Kahneman, D. Thinking, fast and slow. Farrar, Straus and Giroux, 2011. find
  35. Kojima, T., Gu, S. S., Reid, M., Matsuo, Y., and Iwasawa, Y. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35, 2022.
  36. Lai, Y., Zhang, C., Feng, Y., Huang, Q., and Zhao, D. Why machine reading comprehension models learn shortcuts? In Findings of the Association for Computational Linguistics: ACL/IJCNLP 2021, Online Event, August 1-6, 2021, volume ACL/IJCNLP 2021 of Findings of ACL, pp. 989–1002. Association for Comp… find
  37. LeCun, Y. Do large language models need sensory grounding for meaning and understanding? University Lecture, 2024. find
  38. Lee, N., Sreenivasan, K., Lee, J. D., Lee, K., and Papailiopoulos, D. Teaching arithmetic to small transformers. arXiv preprint arXiv:2307.03381, 2023.
  39. Li, X., Trabucco, B., Park, D. H., Luo, M., Shen, S., Darrell, T., and Gao, Y. Discovering non-monotonic autoregressive orderings with variational inference. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. find
  40. Open-Review.net, 2021. URL https://openreview.net/forum?id=jP1vTH3inC. ↗
  41. Li, Y., Huang, Y., Ildiz, M. E., Rawat, A. S., and Oymak, S. Mechanics of next token prediction with self-attention. In 27th International Conference on Artificial Intelligence and Statistics (AISTATS), 2024. find
  42. Lin, C., Jaech, A., Li, X., Gormley, M. R., and Eisner, J. Limitations of autoregressive models and their alternatives. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2021, Online, June 6-11… find
  43. Ling, W., Yogatama, D., Dyer, C., and Blunsom, P. Program induction by rationale generation: Learning to solve and explain algebraic word problems. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017, Vancouver, Canada, July 30 - August 4, Volume 1: L… find
  44. Liu, B., Ash, J. T., Goel, S., Krishnamurthy, A., and Zhang, C. Transformers learn shortcuts to automata. In The Eleventh International Conference on Learning Representations, ICLR 2023, 2023. find
  45. Liu, F. et al. Learning to summarize from human feedback. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020. find
  46. Loshchilov, I. and Hutter, F. Decoupled weight decay regularization. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. find
  47. Lv, A., Zhang, K., Xie, S., Tu, Q., Chen, Y., Wen, J.-R., and Yan, R. Are we falling in a middle-intelligence trap? an analysis and mitigation of the reversal curse. arXiv preprint arXiv:2311.07468, 2023. find
  48. Madaan, A., Tandon, N., Gupta, P., Hallinan, S., Gao, L., Wiegreffe, S., Alon, U., Dziri, N., Prabhumoye, S., Yang, Y., et al. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36, 2024. find
  49. Malach, E. Auto-regressive next-token predictors are universal learners. arXiv preprint arXiv:2309.06979, 2023. find
  50. McCoy, R. T., Yao, S., Friedman, D., Hardy, M., and Griffiths, T. L. Embers of autoregression: Understanding large language models through the problem they are trained to solve. arXiv preprint arXiv:2309.13638, 2023. find
  51. Meng, K., Bau, D., Andonian, A., and Belinkov, Y. Locating and editing factual associations in GPT. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, 2022. find
  52. Merrill, W. and Sabharwal, A. The parallelism tradeoff: Limitations of log-precision transformers, 2023. find
  53. Merrill, W. and Sabharwal, A. The expressive power of transformers with chain of thought. In The Twelfth International Conference on Learning Representations, 2024. find
  54. Momennejad, I., Hasanbeig, H., Frujeri, F. V., Sharma, H., Ness, R. O., Jojic, N., Palangi, H., and Larson, J. Evaluating cognitive maps and planning in large language models with cogeval. Advances in Neural Information Processing Systems, 36, 2023. find
  55. Monea, G., Joulin, A., and Grave, E. Pass: Parallel speculative sampling. 3rd Workshop on Efficient Natural Language and Speech Processing (NeurIPS 2023), 2023. find
  56. Nye, M. I., Andreassen, A. J., Gur-Ari, G., Michalewski, H., Austin, J., Bieber, D., Dohan, D., Lewkowycz, A., Bosma, M., Luan, D., Sutton, C., and Odena, A. Show your work: Scratchpads for intermediate computation with language models. arXiv preprint arXiv:2112.00114, 2021.
  57. Olsson, C., Elhage, N., Nanda, N., Joseph, N., DasSarma, N., Henighan, T., Mann, B., Askell, A., Bai, Y., Chen, A., et al. In-context learning and induction heads. arXiv preprint arXiv:2209.11895, 2022.
  58. Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P. F., Leike, J., and Lowe, R. Training language models to follow instructions w… — Training language models to follow instructions with human feedback
  59. Pal, K., Sun, J., Yuan, A., Wallace, B. C., and Bau, D. Future lens: Anticipating subsequent tokens from a single hidden state. In Proceedings of the 27th Conference on Computational Natural Language Learning, CoNLL 2023. Association for Computational Linguistics, 2023.
  60. Papadopoulos, V., Wenger, J., and Hongler, C. Arrows of time for large language models, 2024. find
  61. Paulus, R., Xiong, C., and Socher, R. A deep reinforced model for abstractive summarization. In 6th International Conference on Learning Representations, ICLR 2018, Conference Track Proceedings.
  62. Pezeshki, M., Kaba, S., Bengio, Y., Courville, A. C., Precup, D., and Lajoie, G. Gradient starvation: A learning proclivity in neural networks. In Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021,… find
  63. Pfau, J., Infanger, A., Sheshadri, A., Panda, A., Michael, J., and Huebner, C. Eliciting language model behaviors using reverse language models. In Socially Responsible Language Modelling Research, 2023. find
  64. Pfungst, O. and Rahn, C. L. Clever Hans (the horse of Mr. Von Osten) a contribution to experimental animal and human psychology. New York, H. Holt and company, 1911. find
  65. Piekos, P., Malinowski, M., and Michalewski, H. Measuring and improving bert’s mathematical abilities by predicting the order of reasoning. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Proce… find
  66. Power, A., Burda, Y., Edwards, H., Babuschkin, I., and Misra, V. Grokking: Generalization beyond overfitting on small algorithmic datasets, 2022.
  67. Qi, W., Yan, Y., Gong, Y., Liu, D., Duan, N., Chen, J., Zhang, R., and Zhou, M. Prophetnet: Predicting future n-gram for sequence-to-sequence pre-training. In Findings of the Association for Computational Linguistics: EMNLP 2020, Online Event, 16-20 November 2020, volume EMNLP 2020 of Findings of AC… find
  68. Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., and Sutskever, I. Language models are unsupervised multitask learners. 2019. find
  69. Ranaldi, L. and Zanzotto, F. M. Hans, are you clever? clever hans effect analysis of neural systems, 2023. find
  70. Ranzato, M., Chopra, S., Auli, M., and Zaremba, W. Sequence level training with recurrent neural networks. In 4th International Conference on Learning Representations, ICLR 2016, Conference Track Proceedings, 2016.
  71. Recchia, G. Teaching autoregressive language models complex tasks by demonstration. arXiv preprint arXiv:2109.02102, 2021. — [arXiv:2109.02102](https://arxiv.org/abs/2109.02102) find
  72. Reynolds, L. and McDonell, K. Prompt programming for large language models: Beyond the few-shot paradigm. In Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, 2021.
  73. Ross, S. and Bagnell, D. Efficient reductions for imitation learning. In Teh, Y. W. and Titterington, D. M. (eds.), Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, AISTATS 2010, Chia Laguna Resort, Sardinia, Italy, May 13-15, 2010, volume 9 of JMLR P… find
  74. Ross, S. and Bagnell, J. A. Reinforcement and imitation learning via interactive no-regret learning. abs/1406.5979, 2014. find
  75. Ross, S., Gordon, G. J., and Bagnell, D. A reduction of imitation learning and structured prediction to no-regret online learning. In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, AISTATS 2011, JMLR Proceedings, 2011. find
  76. Sanford, C., Hsu, D., and Telgarsky, M. Transformers, parallel computation, and logarithmic depth. arXiv preprint arXiv:2402.09268, 2024. find
  77. Shah, H., Tamuly, K., Raghunathan, A., Jain, P., and Netrapalli, P. The pitfalls of simplicity bias in neural networks. In Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, 2020. find
  78. Shalev-Shwartz, S. and Shashua, A. On the sample complexity of end-to-end training vs. semantic abstraction training. arXiv preprint arXiv:1604.06915, 2016. — [arXiv:1604.06915](https://arxiv.org/abs/1604.06915) find
  79. Shalev-Shwartz, S., Shamir, O., and Shammah, S. Failures of gradient-based deep learning. In Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Sydney, NSW, Australia, 6-11 August 2017, volume 70 of Proceedings of Machine Learning Research, pp. 3067–3075. find
  80. Shannon, C. E. A mathematical theory of communication. The Bell System Technical Journal, 27(3):379–423, 1948. find
  81. Shannon, C. E. Prediction and entropy of printed english. The Bell System Technical Journal, 30(1):50–64, 1951. find
  82. Shen, R., Bubeck, S., Eldan, R., Lee, Y. T., Li, Y., and Zhang, Y. Positional description matters for transformers arithmetic. arXiv preprint arXiv:2311.14737, 2023.
  83. Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., and Yao, S. Reflexion: Language agents with verbal reinforcement learning, 2023.
  84. Shlegeris, B., Roger, F., Chan, L., and McLean, E. Language models are better than humans at next-token prediction. arXiv preprint arXiv:2212.11281, 2022. — [arXiv:2212.11281](https://arxiv.org/abs/2212.11281) find
  85. Shridhar, K., Stolfo, A., and Sachan, M. Distilling reasoning capabilities into smaller language models. arXiv preprint arXiv:2212.00193, 2022.
  86. Springer, J. M., Kotha, S., Fried, D., Neubig, G., and Raghunathan, A. Repetition improves language model embeddings, 2024.
  87. Stern, M., Shazeer, N., and Uszkoreit, J. Blockwise parallel decoding for deep autoregressive models. In Advances in Neural Information Processing Systems 31: Annual Conference on Neural Information Processing Systems 2018, NeurIPS 2018, December 3-8, 2018, Montréal, Canada, 2018. find
  88. Thrampoulidis, C. Implicit bias of next-token prediction, 2024. find
  89. Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023.
  90. Tschannen, M., Kumar, M., Steiner, A., Zhai, X., Houlsby, N., and Beyer, L. Image captioners are scalable vision learners too. In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 1… find
  91. Valmeekam, K., Marquez, M., and Kambhampati, S. Can large language models really improve by self-critiquing their own plans? arXiv preprint arXiv:2310.08118, 2023a. — [arXiv:2310.08118](https://arxiv.org/abs/2310.08118) find
  92. Valmeekam, K., Marquez, M., Olmo, A., Sreedharan, S., and Kambhampati, S. Planbench: An extensible benchmark for evaluating large language models on planning and reasoning about change, 2023b.
  93. Valmeekam, K., Marquez, M., Sreedharan, S., and Kambhampati, S. On the planning abilities of large language models - A critical investigation. In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA,…
  94. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., and Polosukhin, I. Attention is all you need. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA, p…
  95. Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E. H., Le, Q. V., and Zhou, D. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, Neur…
  96. Welleck, S., Kulikov, I., Kim, J., Pang, R. Y., and Cho, K. Consistency of a recurrent language model with respect to incomplete decoding. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), November 2020. find
  97. Wies, N., Levine, Y., and Shashua, A. Sub-task decomposition enables learning in sequence to sequence tasks. In The Eleventh International Conference on Learning Representations, ICLR 2023, 2023. find
  98. Williams, R. J. and Zipser, D. A learning algorithm for continually running fully recurrent neural networks. Neural Computation, 1(2):270–280, 1989. find
  99. Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M., Macherey, W., Krikun, M., Cao, Y., Gao, Q., Macherey, K., et al. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144, 2016. find
  100. Xue, F., Likhosherstov, V., Arnab, A., Houlsby, N., Dehghani, M., and You, Y. Adaptive computation with elastic input sequence. In International Conference on Machine Learning, ICML 2023, Proceedings of Machine Learning Research. find
  101. Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T., Cao, Y., and Narasimhan, K. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36, 2023a. find
  102. Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K. R., and Cao, Y. React: Synergizing reasoning and acting in language models. In The Eleventh International Conference on Learning Representations, ICLR 2023, 2023b.
  103. Young, T. and You, Y. On the inconsistencies of conditionals learned by masked language models. arXiv preprint arXiv:2301.00068, 2022. — [arXiv:2301.00068](https://arxiv.org/abs/2301.00068) find
  104. Zelikman, E., Wu, Y., Mu, J., and Goodman, N. D. Star: Bootstrapping reasoning with reasoning. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022, 2022.
  105. Zhang, H., Li, L. H., Meng, T., Chang, K., and den Broeck, G. V. On the paradox of learning to reason from data. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI 2023, 19th-25th August 2023, Macao, SAR, China, pp. 3365–3373. ijcai.org, 2023. find
  106. Zhao, T. Z., Kumar, V., Levine, S., and Finn, C. Learning fine-grained bimanual manipulation with low-cost hardware. In Robotics: Science and Systems XIX, Daegu, Republic of Korea, July 10-14, 2023, 2023. find
  107. Ziegler, D. M., Stiennon, N., Wu, J., Brown, T. B., Radford, A., Amodei, D., Christiano, P., and Irving, G. Fine-tuning language models from human preferences. arXiv preprint arXiv:1909.08593, 2019. find
  108. A. Limitations 1. Our arguments are empirical and conceptual. We have not provided a formal proof for our arguments. 2. We have also not demonstrated failure for very large models such as Llama2 (Touvron et al., 2023) or Mistral (Jiang et al., 2023). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bachmann-2024-pitfalls-next-token-prediction/