# RESULT — Language Models Resist Alignment: Evidence From Data Compression

## Template fill summary
Filled academic project-page template with ACL 2025 paper content.

## Paper
- **Title:** Language Models Resist Alignment: Evidence From Data Compression
- **Authors:** Jiaming Ji*, Kaile Wang*, Tianyi Qiu*, Boyuan Chen*, Jiayi Zhou*, Changye Li, Hantao Lou, Juntao Dai, Yunhuai Liu, Yaodong Yang†
- **Affiliation:** Institute for Artificial Intelligence, Peking University
- **Venue:** ACL 2025 (Volume 1: Long Papers), pages 23411–23432
- **Project:** https://pku-lm-resist-alignment.github.io
- **arXiv:** https://arxiv.org/abs/2406.06144
- **Local PDF note:** local ACL 2025 paper Language Models Resist Alignment (llms-resist-alignment.pdf)

## TL;DR
We demonstrate the elasticity of post-alignment models, forming resistance to alignment.

## Figures embedded
- `figures/fig1_elasticity.png` — Figure 1: Elasticity of language models — spring analogy linking dataset size to compression-rate change.
- `figures/fig2_resistance_pipeline.png` — Figure 2: Resistance experiment pipeline (forward vs inverse alignment).
- `figures/fig3_rebound_pipeline.png` — Figure 3: Rebound experiment pipeline (positive alignment then negative fine-tuning).
- `figures/fig4_rebound_results.png` — Figure 4: Rebound results — deeper positive alignment → faster drop under negative fine-tuning.
- `figures/fig5_scale_model.png` — Figure 5: Larger models exhibit stronger rebound (Qwen 0.5B / 4B / 7B).
- `figures/fig6_scale_pretrain.png` — Figure 6: More pre-training data amplifies rebound (TinyLlama 2.0T / 2.5T / 3.0T).
- `figures/fig7_algorithm_ablation.png` — Figure 7: Rebound holds across PPO / DPO / KTO / SimPO.
- `figures/table1_resistance_loss.png` — Table 1: Inverse alignment loss consistently lower than forward across models and datasets.

## Citation
```bibtex
@inproceedings{ji2025language,
  title={Language Models Resist Alignment: Evidence From Data Compression},
  author={Ji, Jiaming and Wang, Kaile and Qiu, Tianyi and Chen, Boyuan and Zhou, Jiayi and Li, Changye and Lou, Hantao and Dai, Juntao and Liu, Yunhuai and Yang, Yaodong},
  booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  pages={23411--23432},
  year={2025}
}
```

## Static browse
Open `index.html` in a browser from this folder (relative CSS/JS/figures).

Template: lorenmt/clarity-template. Copied assets + clarity/; filled minimal Clarity page as index.html.
### Static browse check
- `static=yes (Clarity assets + clarity/ CSS/JS; optional CDN MathJax/highlight)`
- Path: `outputs/clarity-template/index.html`