Post

Steering vectors can work as activation oracles! Inspired by @isolyth.dev 's recent work on instruct vectors, I've ported the same approach to create Activation Oracles. Seems to work surprisingly well given the param deficit, nearly matching Activation Oracle LoRA performance.

Eris ⌬

@eriskii.net

22 Apr

One thing, it might be nice to see baseline/completely untrained models for comparison. How good is qwen on its own at this vs vectors?

Luna Nova

@lunnova.dev

I can add some more charts in. IIRC: 0% success rate with yes steering no activations and yes activations no steering at taboo.

22/4/26, 10:59 pm0 reposts2 likes

Luna Nova

@lunnova.dev

22 Apr

Added a new section with this chart.

Eris ⌬

@eriskii.net

22 Apr

Can you also put the lora on there maybe?

Luna Nova

@lunnova.dev

22 Apr

Well this is fun. Looks the LoRA gets 60% on the PQA Y/N with no activations... Updated the post with more details and a guess at how it's doing it. lunnova.dev/articles/ste...