/AI6h ago

David Manheim, ALTER founder, argues AI capabilities are accelerating faster than safety, pointing to a widening gap in evaluations

Seth Lazar warns AI governance is unlikely to keep pace.

--0--
Comments
Seth Lazar#1060
Original post
Seth Lazar@sethlazar#1060inAI

I thought he said he reckoned technical alignment work would benefit from the same speed-up as capabilities work? Governance less so, but I think that was more of a throwaway opinion (that part of the discussion was interesting; struck me as missing the point of governance as human control)

David Manheim is @ MoreOnline@davidmanheim

@sethlazar @Alex_A_Guerrero My primary concern about this argument is that continuous doesn't mean gradual, and @rohinmshah has agreed that the acceleration of capabilities is faster than the acceleration of safety. We see this dynamic in the evaluation gap, which is widening already.

11:58 PM · Jun 3, 2026 · 22 Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
VIEWS23LIKES1
Seth Lazar@sethlazar

@davidmanheim @Alex_A_Guerrero @rohinmshah Ah, I was just referring to what Rohin had said, not the object level question. I'm sceptical about a priori arguments on the latter.

David Manheim is @ MoreOnline@davidmanheim

@sethlazar @Alex_A_Guerrero @rohinmshah Unless the vast majority of new work goes into safety, the claim that technical alignment speeds up exactly as fast as capabilities still means it will have a growing lag behind those capabilities.

And even fixing that doesn't help if safety is just patching problems post-hoc.

6hViews 23Likes 1Bookmarks 0