From 7fc5a19b407bd24162028cfc483392affb2c4399 Mon Sep 17 00:00:00 2001 From: "wassname (Michael J Clark)" <1103714+wassname@users.noreply.github.com> Date: Sun, 7 Jun 2026 07:56:33 +0800 Subject: [PATCH] Update README.md --- README.md | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index b1416c3..eecef8e 100644 --- a/README.md +++ b/README.md @@ -7,8 +7,11 @@ What if you can **steer**, **heal** the steering and repeat untill alignment (** Hypothesis: you can distill a steering vector into LoRA weights and "heal" the incoherency the vector injects by regularising the training (KL to base, or weight decay). Then loop and see what multiple rounds give you. -The crux: KL-to-base penalises all drift, persona shift included. The bet is that incoherency drift is large and erratic while the persona shift is small and systematic, so KL kills the incoherency preferentially. If that's wrong, we just trade persona strength for coherence instead of getting both. -image +The method: +- We steer +- Filter completions +- Train a lora with nll and auxiliary loss `rmse(KL(checkpoint, base))`. Why this? Often divergences live in the tail of the distribution change, so this bounds that tail which we care about. We also tested plain KL and it didn't work as well. +- Repeat ## Experiment