Does Fine-Tuning Undo Activation Steering? Behavioural Recovery Without Weight-Edit Reversal
cs.CL, cs.AI, cs.LG
Submitted: 2026-08-25
Updated: 2026-10-03
Code: https://github.com/p-e-w/heretic
Terminology
Sources
- The Llama 3 Herd of Models
- SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling
- Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- A Simple and Efficient Jailbreak Method Exploiting LLMs' Helpfulness
- Orca: Progressive Learning from Complex Explanation Traces of GPT-4
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- Steering Language Models With Activation Engineering
- Qwen3 Technical Report
- An Embarrassingly Simple Defense Against LLM Abliteration Attacks
- Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering