/AI14h ago

Google DeepMind's Joel Z Leibo argues that solipsistic AI trained on static feedback will fail to cooperate with dynamic human institutions

Isolated feedback models trigger self-undermining behaviors once deployed

--0--
Quote posts
Joel Z Leibo#1847
Reposts
David Pfau#153
Original post
Joel Z Leibo@jzl86#1847inAI

Happy to announce our latest paper. We argue that a solipsistic super intelligence is unlikely to be cooperative. The solipsism of the training process gives rise to self undermining in deployment.

Is humanity now building solipsistic or non-solipsistic AI?

Read to find out

11:55 PM · Jun 3, 2026 · 2.6K Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
RETWEETS2
Joel Z Leibo@jzl86

Happy to announce our latest paper. We argue that a solipsistic super intelligence is unlikely to be cooperative. The solipsism of the training process gives rise to self undermining in deployment.

Is humanity now building solipsistic or non-solipsistic AI?

Read to find out

14hViews 2.6KLikes 14Bookmarks 14
Google DeepMind's Joel Z Leibo argues that solipsistic AI trained on static feedback will fail to cooperate with dynamic human institutions · Digg