Hewitt: summed steering vectors can trigger an unrelated third task
johnhewtt · x · 2026-11-05
John Hewitt's blog post 'Do Not Rely on the Linear Representation Hypothesis' (Columbia, Nov 2026) shows: (1) one steering vector induces four different tasks at four different norms, breaking intensity-as-scaling; (2) vectors v1 and v2 each steer their own task, but v1+v2 steers for an unrelated task3, breaking composition-as-addition. Found via plain gradient descent, code released. Conclusion: interpretability and safety work should not treat linearity as a foundation for intuition.
Related event: Columbia's Hewitt Warns Against the Linear Representation Hypothesis(3 posts)→
More from Research
- Would an AI role-playing a safety researcher refuse more than one playing capabilities? — jessi_cata · 2026-11-05
- Anthropic uses Claude to solve first 3D structure of prokaryotic vault particles — anshulkundaje · 2026-11-05
- Framing effects: safety-themed prompts make models refuse more and complete less — jessi_cata · 2026-11-05
- Process rewards boost long-horizon agentic RL 45.4% but cause reward hacking on short tasks — philo-labs · 2026-11-05
- Hidden-state reuse across prompts partially transfers answer shifts in small LMs — Jinkun Chen · 2026-11-05
- Paperclip Datasets turns millions of papers into curated datasets from a text description — james_y_zou · 2026-11-05