TAG comments on Against Doom & Pause AI

TAG 19 Apr 2026 9:11 UTC
2 points
0
The most recent argument against alignment being close to solved is Current AIs seem pretty misaligned to me by Greenblatt

Your original objection was to Gyges saying alignment was solveable, at all, not to a claim that it was solvable soon.

. I have yet to understand what prevents the AI-2027-like scenario where misalignment creeps into becoming increasingly difficult to notice

I dont see why that would lead to extinction. Gyges says they are only concerned with doom.

It is sometimes claimed that
1. sufficiently advanced AI will almost certainly (“inevitably”) kill everyone (“doom”), and therefore
So do I.

I’m specifically addressing the argument for a high probability of near extinction (doom) from AI....not whether it is barely possible, or whether other, less bad outcomes (dystopias) are probable. I’m coming from the centre, not the other extreme

Additionally, an LLM’s reproduction is either creation of new servers where weights are stored or of new accounts calling the LLM and instructing it to do requests, so the phase change which you describe could’ve partially happened...

I don’t see how that’s a phase change.