/AI1d ago

MIRI's Rob Bensinger says Eliezer Yudkowsky's recursive self-improvement model can span months or years rather than happening instantly

Critics argue bounded takeoff contradicts classic foom frameworks.

3600100
#1671xlr8harder
Original post
xlr8harder@xlr8harder#1671inAI

I was speaking loosely, but the cognitive return on investment framing rarely is open to the idea that there could be diminishing ROI, which is what I mean without limit.

But he has apparently invoked lightning fast schedules, which is what would make it an intelligence "explosion"

https://www.astralcodexten.com/p/yudkowsky-contra-christiano-on-ai?utm_campaign=post&utm_medium=web

Rob Bensinger ⏹️@robbensinger

@xlr8harder @croissanthology @willdepue Yudkowsky's original RSI scenario was about a process that takes "months or years, or days or seconds", and it obviously wasn't a process "without any limit", just one that outstrips our ability to react and adapt. I'm not seeing how one avoids that happening sooner or later.

10:22 PM · Jun 5, 2026 · 31 Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS20
xlr8harder@xlr8harder

@robbensinger @croissanthology @willdepue The latter is not the intelligence explosion / foom frame.

Rob Bensinger ⏹️@robbensinger

@xlr8harder @croissanthology @willdepue "ML systems are friendly by default" at least seems like a coherent scenario to me; "AI can never advance scientific frontiers fast enough to outstrip humans' ability to monitor and understand everything" seems a lot harder to imagine.

23hViews 20Likes 1Bookmarks 0
LIKES2
xlr8harder@xlr8harder

@robbensinger @croissanthology @willdepue Oh I misread, I thought you were denying the extremely short timescales, but you were taking issue with the without limit.

Yes he often acknowledges eventually some kind of eventual physical limit, for example, but I think diminishing returns on investment is more pertinent.

xlr8harder@xlr8harder

I was speaking loosely, but the cognitive return on investment framing rarely is open to the idea that there could be diminishing ROI, which is what I mean without limit.

But he has apparently invoked lightning fast schedules, which is what would make it an intelligence "explosion"

https://www.astralcodexten.com/p/yudkowsky-contra-christiano-on-ai?utm_campaign=post&utm_medium=web

23hViews 15Likes 2Bookmarks 0
REPLIES1
Rob Bensinger ⏹️@robbensinger

@xlr8harder @croissanthology @willdepue "ML systems are friendly by default" at least seems like a coherent scenario to me; "AI can never advance scientific frontiers fast enough to outstrip humans' ability to monitor and understand everything" seems a lot harder to imagine.

1dViews 16