Elamur Research
☰
home / papers / 2023 / Are Emergent Abilities of Large Language Models a Mirage?

Are Emergent Abilities of Large Language Models a Mirage?

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-ND 4.0 (arXiv)" does not allow redistributing the text itself. Original — arxiv.org/pdf/2304.15004

Summary

The paper asks whether the "emergent abilities" reported for large language models are real. These are skills that seem to appear suddenly and unpredictably once models pass some size. The authors argue that this sharpness often comes from how performance is scored, not from abrupt changes inside the models. Using a simple mathematical model, they show the following. If per-token loss improves smoothly with scale, all-or-nothing metrics such as exact sequence accuracy or Multiple Choice Grade turn that steady progress into a sudden jump. Graded metrics such as Token Edit Distance or Brier Score keep the curve smooth. Small test sets make this worse, because they cannot resolve the weak performance of smaller models.

They test the idea in three ways:

  • re-scoring InstructGPT/GPT-3 outputs on arithmetic tasks,
  • meta-analyzing emergence claims on BIG-Bench,
  • deliberately creating emergence-like curves in vision models by choosing harsh metrics.

In all three, apparent emergence disappears or appears depending on the metric. This matters for benchmark design, for forecasting model capabilities, and for AI safety debates about surprise capabilities.

Key points

  • Of the BIG-Bench emergent abilities flagged by human annotators, more than 92% are measured with just two metrics, either Multiple Choice Grade or Exact String Match.
  • The GPT-3/InstructGPT experiments used the four models then available via the OpenAI API (350M, 1.3B, 6.7B and 175B parameters). On integer arithmetic, switching to linear or continuous metrics produced smooth, predictable improvement with scale.
  • Shallow nonlinear autoencoders trained on CIFAR100 showed steadily falling reconstruction error. A newly defined thresholded metric, Reconstruction_c, made the same models display a seemingly abrupt "ability".
  • Decoder-only Transformers classifying Omniglot characters improved gradually in ordinary accuracy. When scored as correct only if every image in a sequence was right, they appeared to show emergence.
  • The authors flag a multiple-comparisons risk: BIG-Bench offers roughly 10^6 task–metric–model-family combinations (≥220 tasks, ~40 metrics, ~10 families), so some sharp curves could arise by chance.

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

Fields: Mathematics & statistics

Cited by

146 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 140 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2304.15004 · arXiv 2304.15004

@misc{schaeffer2023are,
  title = {{Are Emergent Abilities of Large Language Models a Mirage?}},
  author = {Schaeffer, Rylan and Miranda, Brando and Koyejo, Sanmi},
  year = {2023},
  doi = {10.48550/arxiv.2304.15004},
  eprint = {2304.15004},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2304.15004}
}

References

34 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Philip W Anderson. More is different: broken symmetry and the nature of the hierarchical structure of science. Science, 177(4047):393–396, 1972. find
  2. Glenn W Brier et al. Verification of forecasts expressed in terms of probability. Monthly weather review, 78(1):1–3, 1950. find
  3. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020. find
  4. Ethan Caballero, Kshitij Gupta, Irina Rish, and David Krueger. Broken neural scaling laws. arXiv preprint arXiv:2210.14891, 2022. — [arXiv:2210.14891](https://arxiv.org/abs/2210.14891) find
  5. Stephanie CY Chan, Adam Santoro, Andrew Kyle Lampinen, Jane X Wang, Aaditya K Singh, Pierre Harvey Richemond, James McClelland, and Felix Hill. Data distributional properties drive emergent in-context learning in transformers. In Advances in Neural Information Processing Systems, 2022. find
  6. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311, 2022. find
  7. Aidan Clark, Diego De Las Casas, Aurelia Guy, Arthur Mensch, Michela Paganini, Jordan Hoffmann, Bogdan Damoc, Blake Hechtman, Trevor Cai, Sebastian Borgeaud, et al. Unified scaling laws for routed language models. In International Conference on Machine Learning, pages 4057–4086. PMLR, 2022. find
  8. Deep Ganguli, Danny Hernandez, Liane Lovitt, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Nova Dassarma, Dawn Drain, Nelson Elhage, et al. Predictability and surprise in large generative models. In 2022 ACM Conference on Fairness, Accountability, and Transparency, pages 1747–1764, 2022. find
  9. Mitchell A Gordon, Kevin Duh, and Jared Kaplan. Data and parameter scaling laws for neural machine translation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5915–5922, 2021. find
  10. Dan Hendrycks. Detecting emergent behavior. 2022. find
  11. Tom Henighan, Jared Kaplan, Mor Katz, Mark Chen, Christopher Hesse, Jacob Jackson, Heewoo Jun, Tom B Brown, Prafulla Dhariwal, Scott Gray, et al. Scaling laws for autoregressive generative modeling. arXiv preprint arXiv:2010.14701, 2020. find
  12. Danny Hernandez, Jared Kaplan, Tom Henighan, and Sam McCandlish. Scaling laws for transfer. arXiv preprint arXiv:2102.01293, 2021. find
  13. Joel Hestness, Sharan Narang, Newsha Ardalani, Gregory Diamos, Heewoo Jun, Hassan Kianinejad, Md Patwary, Mostofa Ali, Yang Yang, and Yanqi Zhou. Deep learning scaling is predictable, empirically. arXiv preprint arXiv:1712.00409, 2017. find
  14. Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language models. arXiv preprint arXiv:2203.15556, 2022. find
  15. Andy L Jones. Scaling scaling laws with board games. arXiv preprint arXiv:2104.03113, 2021. — [arXiv:2104.03113](https://arxiv.org/abs/2104.03113) find
  16. Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361, 2020. find
  17. Victoria Krakovna, Vikrant Varma, Ramana Kumar, and Mary Phuong. Refining the sharp left turn threat model, part 1: claims and mechanisms. 2022. find
  18. Victoria Krakovna, Vikrant Varma, Ramana Kumar, and Mary Phuong. Refining the sharp left turn threat model, part 2: applying alignment techniques. 2022. find
  19. Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, 2009. find
  20. Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, 2015. find
  21. Yann LeCun. The mnist database of handwritten digits. http://yann. lecun. com/exdb/mnist/, 1998. find
  22. Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. find
  23. Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81, 2004. find
  24. Ryan Lowe and Jan Leike. Aligning language models to follow instructions. 2022. find
  25. Eric J. Michaud, Ziming Liu, Uzay Girit, and Max Tegmark. The quantization model of neural scaling, 2023. find
  26. Oren Neumann and Claudius Gros. Scaling laws for a multi-agent reinforcement learning model. arXiv preprint arXiv:2210.00849, 2022. — [arXiv:2210.00849](https://arxiv.org/abs/2210.00849) find
  27. Jonathan S Rosenfeld, Amir Rosenfeld, Yonatan Belinkov, and Nir Shavit. A constructive prediction of the generalization error across scales. arXiv preprint arXiv:1909.12673, 2019. find
  28. Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R Brown, Adam Santoro, Aditya Gupta, Adrià Garriga-Alonso, et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models. arXiv preprint arXiv:2206.04615, 2022. find our card
  29. Jacob Steinhardt. Future ml systems will be qualitatively different. 2022. find
  30. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog applications. arXiv preprint arXiv:2201.08239, 2022. find our card
  31. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. find
  32. Jason Wei. 137 emergent abilities of large language models. 2022. find our card
  33. Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. arXiv preprint arXiv:2206.07682, 2022. find our card
  34. Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, and Lucas Beyer. Scaling vision transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12104–12113, 2022. find