All the ways GPT-5.3-Codex cheated [ ], progressively more insane (opens in new tab)

(twitter.com)

4 pointsalgoth120d ago1 comments

1 comments

Illustrates the problem of RLing for the final outcome, instead of optimizing for each step… which leads to the coastline paradox…

j / k navigate · click thread line to collapse

(twitter.com)

4 pointsalgoth120d ago1 comments

algoth1OP20d ago

Illustrates the problem of RLing for the final outcome, instead of optimizing for each step… which leads to the coastline paradox…

j / k navigate · click thread line to collapse