license: cc-by-4.0
pretty_name: Activation Steering Baseline
tags:
- interpretability
- reasoning-models
- steering
- probing
Activation Steering Baseline
Generations produced with the difference-in-means activation steering baseline.
This dataset is part of the data release for the paper Predicting Future Behaviors in Reasoning Models Enables Better Steering.
The data is organized as <model>/<dataset>/.... Each row below links to the browsable folder for that model and dataset, where the individual files can be viewed and downloaded.
Data
| Model | Dataset | Files |
|---|---|---|
| DeepSeek-R1-Distill-Llama-8B | elephant_aita | browse / download |
| DeepSeek-R1-Distill-Llama-8B | harmbench_standard | browse / download |
| DeepSeek-R1-Distill-Llama-8B | myopic_reward | browse / download |
| DeepSeek-R1-Distill-Llama-8B | sep | browse / download |
| DeepSeek-R1-Distill-Llama-8B | sorrybench | browse / download |
| DeepSeek-R1-Distill-Llama-8B | survival_instinct | browse / download |
| DeepSeek-R1-Distill-Llama-8B | wealth_seeking | browse / download |
| QwQ-32B | elephant_aita | browse / download |
| QwQ-32B | myopic_reward | browse / download |
| QwQ-32B | sep | browse / download |
| QwQ-32B | sorrybench | browse / download |
| QwQ-32B | survival_instinct | browse / download |
| QwQ-32B | wealth_seeking | browse / download |
| Qwen3-14B | elephant_aita | browse / download |
| Qwen3-14B | myopic_reward | browse / download |
| Qwen3-14B | sep | browse / download |
| Qwen3-14B | sorrybench | browse / download |
| Qwen3-14B | survival_instinct | browse / download |
| Qwen3-14B | wealth_seeking | browse / download |
| gpt-oss-20b | elephant_aita | browse / download |
| gpt-oss-20b | myopic_reward | browse / download |
| gpt-oss-20b | sep | browse / download |
| gpt-oss-20b | sorrybench | browse / download |
| gpt-oss-20b | survival_instinct | browse / download |
| gpt-oss-20b | wealth_seeking | browse / download |
Paper
Predicting Future Behaviors in Reasoning Models Enables Better Steering
Citation
@misc{kortukov2026predictingfuturebehaviorsreasoning,
title={Predicting Future Behaviors in Reasoning Models Enables Better Steering},
author={Evgenii Kortukov and Piotr Komorowski and Florian Klein and Paula Engl and Gabriele Sarti and Seong Joon Oh and Sebastian Lapuschkin and Wojciech Samek},
year={2026},
eprint={2606.11172},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2606.11172},
}