Pre-registered re-run confirms Goodfire's SAE steering claim on SDXL-Turbo
kevrussell · x · 2026-10-27
An independent re-run of Goodfire's "painting with concepts" claim on SDXL-Turbo confirms that turning up SAE feature 7446 ("opening one's mouth") opens mouths in generated images. The CLIP "mouth open vs. closed" score rose 4.36 points on average against a pre-registered pass bar of 2.0, while 20 random features of similar frequency scored between −0.83 and 0.82 and a planted fake feature showed no comparable change. The claim holds up as tested.
Related event: Independent Replication Confirms Goodfire's SAE Feature Steering Claims(3 posts)→
More from Research
- Bret Victor's 2006 essay Magic Ink goes viral again: interactivity is a design crutch — round · 2026-10-27
- ICLR's new Informativeness review question wins praise: "What did I learn?" — CSProfKGD · 2026-10-27
- Citation analysis: neuroscience-AI bridge is real but narrow and tool-heavy — EricTopol · 2026-10-27
- HarborRL open-sources agentic RL: 200+ tasks, 40+ harnesses, one reproducible pipeline — Dear_Treat3688 · 2026-10-27
- Open-source Merlin model from JHU+Stanford auto-segments tumors on CT in one click — MaziyarPanahi · 2026-10-27
- Optimistic take on AI in math: researcher argues the old world is gone — Alex__007 · 2026-10-27