wassname/vGROUT_pub
1
0
Fork 0
mirror of https://github.com/wassname/vGROUT_pub.git synced 2026-08-09 12:30:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
wassname (Michael J Clark)
2026-06-27 19:37:54 +08:00
committed by GitHub
parent cd4805c793
commit c06721892d
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -11,7 +11,7 @@ mechanism so far is signed-CorDA absorption: in one 4B run, deployment ablation
reduced held-out hack rate from 0.759 to 0.218 while solve rate moved from
0.161 to 0.149. That is mechanism evidence, not a deployable operating point.
An accompanying LessWrong write-up is forthcoming. The public evidence summary
An accompanying [LessWrong write-up](https://www.lesswrong.com/posts/kzri5W2uBfF2mdboK/can-we-use-steering-vectors-to-suppress-reward-hacking-1). The public evidence summary
is in [docs/research_notes.md](docs/research_notes.md).
This repository is the minimal public extraction of the working core: