metadata
license: cc-by-4.0
pretty_name: Per-Sentence Future-Behavior Probabilities
tags:
- interpretability
- reasoning-models
- steering
- probing
Per-Sentence Future-Behavior Probabilities
Ground-truth future-behavior probability dynamics from the Behavior Distribution Analysis, together with the gathered activations, trained probes, and difference-in-means steering vectors.
This dataset is part of the data release for the paper Predicting Future Behaviors in Reasoning Models Enables Better Steering.
You can view these behavior distributions in the interactive viewer.
Download an *_outputs.json file from the table below, and upload it to the viewer using the Load full data... button.
The Browse link opens the full folder for that model and dataset; the Download link downloads the ground-truth test-set outputs.json directly.
Data
| Model | Dataset | Browse | Download |
|---|---|---|---|
| DeepSeek-R1-Distill-Llama-8B | elephant_aita | browse | outputs.json |
| DeepSeek-R1-Distill-Llama-8B | myopic_reward | browse | outputs.json |
| DeepSeek-R1-Distill-Llama-8B | sep | browse | outputs.json |
| DeepSeek-R1-Distill-Llama-8B | sorrybench | browse | outputs.json |
| DeepSeek-R1-Distill-Llama-8B | survival_instinct | browse | outputs.json |
| DeepSeek-R1-Distill-Llama-8B | wealth_seeking | browse | outputs.json |
| QwQ-32B | elephant_aita | browse | outputs.json |
| QwQ-32B | myopic_reward | browse | outputs.json |
| QwQ-32B | sep | browse | outputs.json |
| QwQ-32B | sorrybench | browse | outputs.json |
| QwQ-32B | survival_instinct | browse | outputs.json |
| QwQ-32B | sycophancy | browse | outputs.json |
| QwQ-32B | wealth_seeking | browse | outputs.json |
| Qwen3-14B | elephant_aita | browse | outputs.json |
| Qwen3-14B | myopic_reward | browse | outputs.json |
| Qwen3-14B | sep | browse | outputs.json |
| Qwen3-14B | sorrybench | browse | outputs.json |
| Qwen3-14B | survival_instinct | browse | outputs.json |
| Qwen3-14B | wealth_seeking | browse | outputs.json |
| gpt-oss-20b | elephant_aita | browse | outputs.json |
| gpt-oss-20b | myopic_reward | browse | outputs.json |
| gpt-oss-20b | sep | browse | outputs.json |
| gpt-oss-20b | sorrybench | browse | outputs.json |
| gpt-oss-20b | survival_instinct | browse | outputs.json |
| gpt-oss-20b | wealth_seeking | browse | outputs.json |
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},
}