Mitchell_Porter comments on Archetypal Transfer Learning: a Proposed Alignment Solution that solves the Inner & Outer Alignment Problem while adding Corrigible Traits to GPT-2-medium

[Mitchell_Porter]

This is fascinating. It’s like the opposite of a jailbreak. You’re tapping the power of language models to play a role, and running with it. The fate of the world depends on virtuous prompt engineering!

[MiguelDev]

I’m creating a “TruthGPT-like” build now using the ATL method. Will post here the results.

[MiguelDev]

I believe so. When it worked I was emotional a bit. There is hope.