Dynamic Abliteration: Non-Destructive Refusal Suppression via Engram Steering
Thread
Unofficial Hacker News client; not affiliated with Y Combinator.
Dynamic Abliteration: Non-Destructive Refusal Suppression via Engram Steering
Unofficial Hacker News client; not affiliated with Y Combinator.
qgin · · focus · HN ↗
We don't even know how to align models, but even if we did, apparently undoing that alignment if trivial.
Really I'm looking for any argument that lays out a scenario where this works out.
cyanydeez · · focus · HN ↗
ChickeNES · · focus · HN ↗
cyanydeez · · focus · HN ↗
ChickeNES · · focus · HN ↗