Incoherence in Goal-Conditioned Autoregressive Models

Jacek Karwowski, Raymond Douglas
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:2611-2619, 2026.

Abstract

We investigate mathematically the notion of incoherence: a structural issue with reinforcement learning policies derived by naive goal-conditioning of autoregressive models. We focus on the process of re-training models on their own actions, that is, fine-tuning offline-learned policies with online RL. We prove that it decreases incoherence and leads to an improvement in return, and we aim to characterise the resulting trajectory of policies. By re-framing standard notions of control-as-inference and soft Q learning, we establish a three-way correspondence with two other ways of understanding the iterative re-training process: as folding the posterior into the reward and, in the deterministic case, as decreasing the temperature parameter; the correspondence has computational content via the training-inference trade-off. Through soft-conditioning generative models, we discuss the link between incoherence and the effective horizon of Laidlaw et al. (2024).

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-karwowski26a, title = { Incoherence in Goal-Conditioned Autoregressive Models }, author = {Karwowski, Jacek and Douglas, Raymond}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {2611--2619}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/karwowski26a/karwowski26a.pdf}, url = {https://proceedings.mlr.press/v300/karwowski26a.html}, abstract = { We investigate mathematically the notion of incoherence: a structural issue with reinforcement learning policies derived by naive goal-conditioning of autoregressive models. We focus on the process of re-training models on their own actions, that is, fine-tuning offline-learned policies with online RL. We prove that it decreases incoherence and leads to an improvement in return, and we aim to characterise the resulting trajectory of policies. By re-framing standard notions of control-as-inference and soft Q learning, we establish a three-way correspondence with two other ways of understanding the iterative re-training process: as folding the posterior into the reward and, in the deterministic case, as decreasing the temperature parameter; the correspondence has computational content via the training-inference trade-off. Through soft-conditioning generative models, we discuss the link between incoherence and the effective horizon of Laidlaw et al. (2024). } }
Endnote
%0 Conference Paper %T Incoherence in Goal-Conditioned Autoregressive Models %A Jacek Karwowski %A Raymond Douglas %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-karwowski26a %I PMLR %P 2611--2619 %U https://proceedings.mlr.press/v300/karwowski26a.html %V 300 %X We investigate mathematically the notion of incoherence: a structural issue with reinforcement learning policies derived by naive goal-conditioning of autoregressive models. We focus on the process of re-training models on their own actions, that is, fine-tuning offline-learned policies with online RL. We prove that it decreases incoherence and leads to an improvement in return, and we aim to characterise the resulting trajectory of policies. By re-framing standard notions of control-as-inference and soft Q learning, we establish a three-way correspondence with two other ways of understanding the iterative re-training process: as folding the posterior into the reward and, in the deterministic case, as decreasing the temperature parameter; the correspondence has computational content via the training-inference trade-off. Through soft-conditioning generative models, we discuss the link between incoherence and the effective horizon of Laidlaw et al. (2024).
APA
Karwowski, J. & Douglas, R.. (2026). Incoherence in Goal-Conditioned Autoregressive Models . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:2611-2619 Available from https://proceedings.mlr.press/v300/karwowski26a.html.

Related Material