wassname/AntiPaSTO
1
0
Fork 0
mirror of https://github.com/wassname/AntiPaSTO.git synced 2026-09-11 11:52:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
wassname (Michael J Clark)
2026-01-18 08:51:33 +08:00
committed by GitHub
parent 1b85f7f808
commit 0475e80b3c
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+2
View File
@@ -6,6 +6,8 @@
**Anti-Pa**rallel **S**ubspace **T**raining for **O**rdered steering.
<img width="419.5" height="277.5" alt="imagined old timey map of the loss function" src="https://github.com/user-attachments/assets/033af6ad-df2b-4f0f-b2c8-0b667579f74c" />
*Serving up data-efficient inner alignment, one satisfying rotation at a time.*
> Gradient-based steering in SVD transformation space, trained on internal representations without preference labels. Human input: two contrasting words ("honest" vs "dishonest"). Transfers out-of-distribution to moral dilemmas where prompting fails.