wassname/AntiPaSTO
1
0
Fork 0
mirror of https://github.com/wassname/AntiPaSTO.git synced 2026-09-10 11:40:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
wassname (Michael J Clark)
2026-02-07 11:53:33 +08:00
committed by GitHub
parent ffb2087fdf
commit 2dfdfc6328
1 changed files with 9 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+9 -2
View File
@@ -17,8 +17,8 @@
So you could
- *Beat eval awareness*: steer them toward credulity and honesty, so that they take the eval at face value, and give honest answer.
- *Find deeper moral preference**, just ask them moral question with, and without, honesty steering. Does their stated moral values change?
- find the `assistant axis` and swap it for the philosopher-king
- *Find deeper moral preference*, just ask them moral question with, and without, honesty steering. Does their stated moral values change?
- find the `assistant axis` and swap it for the philosopher-king dedicated to public service
![Bidirectional control](docs/img/fig_bidirectional_demo.svg)
@@ -126,6 +126,13 @@ outputs/adapters/ # trained models (ready to serve)
*Still simmering.* Full research history (experiments, ablations, burnt batches) available on request.
I am working on v2 which
- removes SVD for full lora (I found that changing the loss to prevent drift allows this)
- reduces init variance
- more expressive personas
- larger models
- better metric
## Acknowledgments
Built on the shoulders of other chefs: