Strategy Coopetition Explains the Emergence and Transience of In-Context Learning

Aaditya K Singh, Ted Moskovitz, Sara Dragutinović, Felix Hill, Stephanie C.Y. Chan, Andrew M Saxe
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:55720-55739, 2025.

Abstract

In-context learning (ICL) is a powerful ability that emerges in transformer models, enabling them to learn from context without weight updates. Recent work has established emergent ICL as a transient phenomenon that can sometimes disappear after long training times. In this work, we sought a mechanistic understanding of these transient dynamics. Firstly, we find that—after the disappearance of ICL—the asymptotic strategy is a remarkable hybrid between in-weights and in-context learning, which we term “context-constrained in-weights learning” (CIWL). CIWL is in competition with ICL, and eventually replaces it as the dominant strategy of the model (thus leading to ICL transience). However, we also find that the two competing strategies actually share sub-circuits, which gives rise to cooperative dynamics as well. For example, in our setup, ICL is unable to emerge quickly on its own, and can only be enabled through the simultaneous slow development of asymptotic CIWL. CIWL thus both cooperates and competes with ICL, a phenomenon we term “strategy coopetition”. We propose a minimal mathematical model that reproduces these key dynamics and interactions. Informed by this model, we were able to identify a setup where ICL is truly emergent and persistent.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-singh25c, title = {Strategy Coopetition Explains the Emergence and Transience of In-Context Learning}, author = {Singh, Aaditya K and Moskovitz, Ted and Dragutinovi\'{c}, Sara and Hill, Felix and Chan, Stephanie C.Y. and Saxe, Andrew M}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {55720--55739}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/singh25c/singh25c.pdf}, url = {https://proceedings.mlr.press/v267/singh25c.html}, abstract = {In-context learning (ICL) is a powerful ability that emerges in transformer models, enabling them to learn from context without weight updates. Recent work has established emergent ICL as a transient phenomenon that can sometimes disappear after long training times. In this work, we sought a mechanistic understanding of these transient dynamics. Firstly, we find that—after the disappearance of ICL—the asymptotic strategy is a remarkable hybrid between in-weights and in-context learning, which we term “context-constrained in-weights learning” (CIWL). CIWL is in competition with ICL, and eventually replaces it as the dominant strategy of the model (thus leading to ICL transience). However, we also find that the two competing strategies actually share sub-circuits, which gives rise to cooperative dynamics as well. For example, in our setup, ICL is unable to emerge quickly on its own, and can only be enabled through the simultaneous slow development of asymptotic CIWL. CIWL thus both cooperates and competes with ICL, a phenomenon we term “strategy coopetition”. We propose a minimal mathematical model that reproduces these key dynamics and interactions. Informed by this model, we were able to identify a setup where ICL is truly emergent and persistent.} }
Endnote
%0 Conference Paper %T Strategy Coopetition Explains the Emergence and Transience of In-Context Learning %A Aaditya K Singh %A Ted Moskovitz %A Sara Dragutinović %A Felix Hill %A Stephanie C.Y. Chan %A Andrew M Saxe %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-singh25c %I PMLR %P 55720--55739 %U https://proceedings.mlr.press/v267/singh25c.html %V 267 %X In-context learning (ICL) is a powerful ability that emerges in transformer models, enabling them to learn from context without weight updates. Recent work has established emergent ICL as a transient phenomenon that can sometimes disappear after long training times. In this work, we sought a mechanistic understanding of these transient dynamics. Firstly, we find that—after the disappearance of ICL—the asymptotic strategy is a remarkable hybrid between in-weights and in-context learning, which we term “context-constrained in-weights learning” (CIWL). CIWL is in competition with ICL, and eventually replaces it as the dominant strategy of the model (thus leading to ICL transience). However, we also find that the two competing strategies actually share sub-circuits, which gives rise to cooperative dynamics as well. For example, in our setup, ICL is unable to emerge quickly on its own, and can only be enabled through the simultaneous slow development of asymptotic CIWL. CIWL thus both cooperates and competes with ICL, a phenomenon we term “strategy coopetition”. We propose a minimal mathematical model that reproduces these key dynamics and interactions. Informed by this model, we were able to identify a setup where ICL is truly emergent and persistent.
APA
Singh, A.K., Moskovitz, T., Dragutinović, S., Hill, F., Chan, S.C. & Saxe, A.M.. (2025). Strategy Coopetition Explains the Emergence and Transience of In-Context Learning. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:55720-55739 Available from https://proceedings.mlr.press/v267/singh25c.html.

Related Material