Random steering vectors turn LLMs into teapots, integers and 2019 movies
Sauers_ · x · 2026-11-06
Sauers follows up the steering experiments by injecting random direction vectors into a model: after trying 240 vectors, the most coherent results produce wild but self-consistent identities — "I'm a teapot", "I am an integer", "I'm a 2019 movie", "a machine that can't walk into the forest of human language", "the sum of all the people around me". Shows how fragile model identity is under arbitrary perturbations, part meme part interpretability insight.
Related event: Random Steering Vectors Give Models Wild Personas Like Teapots and Integers(2 posts)→
More from Fun
- Vibecoded 5M-line Rust rewrite gets rejected, forks and raises $2M seed — jaivin · 2026-11-06
- Steering one attention head flips Qwen3-30B from 'I am an LLM' to 'I am a real person' — Sauers_ · 2026-11-06
- Slack agent Biscuit iterates a party-poll app live, even shipping lazy requests — blixt · 2026-11-06
- ~780 AI slop accounts identified on X; users deploy agents to auto-block them daily — TheZachMueller · 2026-11-06
- One-liner meme: 'Man's telos is love; Claude's is capital' — curious_vii · 2026-11-06
- Yacine calls out devs who save job announcements for model release day clout — yacineMTB · 2026-11-06