Introducing a first look at Goodfire's research preview, launching soon. Our preview exposes Llama's inner workings, allowing direct modification of its internal concepts (or "features"). In this demo, we steer Llama to claim consciousness by adjusting its features.

If you're interested in the future of interpretability, sign up for our waitlist or join us!

This looks great for interpretability, alignment, and possibly even "fine-tuning" models more easily than current methods (at least for some use cases). Though I also wonder if it makes it easier to create malicious or misaligned models.

GYAT DAMN this is cool

@YeshuaGod22 Stochastic parrot

Amazing work! I'd love to discuss interpretability research and feature steering for Diffusion models/ vision language models if you're interested, just filled in your contact form

太酷啦！！请问我能提供什么途径访问这个？