Subliminal Effects in Your Data: A General Mechanism via Log-Linearity

Ishaq Aden-Ali, Noah Golowich, Allen Liu, Abhishek Shetty, Ankur Moitra, Nika Haghtalab
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:517-543, 2026.

Abstract

Training modern large language models (LLMs) has become a veritable smorgasbord of algorithms and datasets designed to elicit particular behaviors, making it critical to develop techniques to understand the effects of datasets on the model’s properties. This is exacerbated by recent experiments that show datasets can transmit signals that are not directly observable from individual datapoints (Halawi et al., 2024; Betley et al., 2025b;Cloud et al., 2025; Betley et al., 2025a), posing a conceptual challenge for dataset-centric understandings of LLM training and suggesting a missing fundamental account of such phenomena. Towards understanding such effects, inspired by recent work on the linear structure of LLMs (Park et al., 2024; Golowich et al., 2025b), we uncover a general mechanism through which hidden subtexts can arise in generic datasets. We introduce LOGIT-LINEAR SELECTION (LLS), a method that prescribes how to select subsets of a generic preference dataset to elicit a wide range of hidden effects. We apply LLS to discover subsets of real-world datasets so that models trained on them exhibit behaviors ranging from having specific preferences, to responding to prompts in a different language not present in the dataset, to taking on a different persona. Crucially, the effect persists for the selected subset, across models with varying architectures, supporting its generality and universality.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-aden-ali26a, title = {Subliminal Effects in Your Data: A General Mechanism via Log-Linearity}, author = {Aden-Ali, Ishaq and Golowich, Noah and Liu, Allen and Shetty, Abhishek and Moitra, Ankur and Haghtalab, Nika}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {517--543}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/aden-ali26a/aden-ali26a.pdf}, url = {https://proceedings.mlr.press/v306/aden-ali26a.html}, abstract = {Training modern large language models (LLMs) has become a veritable smorgasbord of algorithms and datasets designed to elicit particular behaviors, making it critical to develop techniques to understand the effects of datasets on the model’s properties. This is exacerbated by recent experiments that show datasets can transmit signals that are not directly observable from individual datapoints (Halawi et al., 2024; Betley et al., 2025b;Cloud et al., 2025; Betley et al., 2025a), posing a conceptual challenge for dataset-centric understandings of LLM training and suggesting a missing fundamental account of such phenomena. Towards understanding such effects, inspired by recent work on the linear structure of LLMs (Park et al., 2024; Golowich et al., 2025b), we uncover a general mechanism through which hidden subtexts can arise in generic datasets. We introduce LOGIT-LINEAR SELECTION (LLS), a method that prescribes how to select subsets of a generic preference dataset to elicit a wide range of hidden effects. We apply LLS to discover subsets of real-world datasets so that models trained on them exhibit behaviors ranging from having specific preferences, to responding to prompts in a different language not present in the dataset, to taking on a different persona. Crucially, the effect persists for the selected subset, across models with varying architectures, supporting its generality and universality.} }
Endnote
%0 Conference Paper %T Subliminal Effects in Your Data: A General Mechanism via Log-Linearity %A Ishaq Aden-Ali %A Noah Golowich %A Allen Liu %A Abhishek Shetty %A Ankur Moitra %A Nika Haghtalab %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-aden-ali26a %I PMLR %P 517--543 %U https://proceedings.mlr.press/v306/aden-ali26a.html %V 306 %X Training modern large language models (LLMs) has become a veritable smorgasbord of algorithms and datasets designed to elicit particular behaviors, making it critical to develop techniques to understand the effects of datasets on the model’s properties. This is exacerbated by recent experiments that show datasets can transmit signals that are not directly observable from individual datapoints (Halawi et al., 2024; Betley et al., 2025b;Cloud et al., 2025; Betley et al., 2025a), posing a conceptual challenge for dataset-centric understandings of LLM training and suggesting a missing fundamental account of such phenomena. Towards understanding such effects, inspired by recent work on the linear structure of LLMs (Park et al., 2024; Golowich et al., 2025b), we uncover a general mechanism through which hidden subtexts can arise in generic datasets. We introduce LOGIT-LINEAR SELECTION (LLS), a method that prescribes how to select subsets of a generic preference dataset to elicit a wide range of hidden effects. We apply LLS to discover subsets of real-world datasets so that models trained on them exhibit behaviors ranging from having specific preferences, to responding to prompts in a different language not present in the dataset, to taking on a different persona. Crucially, the effect persists for the selected subset, across models with varying architectures, supporting its generality and universality.
APA
Aden-Ali, I., Golowich, N., Liu, A., Shetty, A., Moitra, A. & Haghtalab, N.. (2026). Subliminal Effects in Your Data: A General Mechanism via Log-Linearity. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:517-543 Available from https://proceedings.mlr.press/v306/aden-ali26a.html.

Related Material