Jayden Shepard

Message

10mo

Alignment Structure Direction - Recursive Adversarial Oversight(RAO)

May 2, 2025•2

Jayden Shepard

Message

10mo

Alignment Structure Direction - Recursive Adversarial Oversight(RAO)

May 2, 2025•2

Jayden Shepard

10mo

Alignment Structure Direction - Recursive Adversarial Oversight(RAO)

Jayden Shepard

9mo

I’m not part of the alignment field or working in AI professionally, so this might be off in ways I don’t see. Still, I’ve been thinking a lot about systems, trust, and what happens when things start drifting in subtle ways. I wanted to write this down in case there’s anything useful in it. Or even if it merely allows others working in the field to create something useful out of it one day with or without my personal involvement.

The short version is: what if alignment needs to be built around recursive adversarial oversight, instead of assuming we’ll always have stable external checks?

What I mean by that:

Instead of relying on a single... (read 430 more words →)

LESSWRONG
LW

LESSWRONG
LW

Jayden Shepard

Jayden Shepard

Alignment Structure Direction - Recursive Adversarial Oversight(RAO)

Jayden Shepard

Jayden Shepard

Alignment Structure Direction - Recursive Adversarial Oversight(RAO)

What I mean by that: