Founder, agent infrastructure and interfaces at Fulcrum Research, https://fulcrum.inc
Past: MIT, MAIA, theory of deep learning research
More info: https://uzpg.me
zef
Karma: 179
nice, i tried this experiment a few years ago and it failed, nice to see that it works now! spooky stuff
John Wentworth has nice content related to showing these sorts of inner models.
I enjoyed this post. It’s very thorough and develops something I think people are not thinking about enough, often bundling capabilities all in one bunch without considering what kind of capabilities are differentially useful and unsatisfied by the market when it comes to this kind of iterated deference world.