AI ALIGNMENT FORUM
AF

wkey22
Ω5000
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No wikitag contributions to display.
Why GPT wants to mesa-optimize & how we might change this
wkey225y*50

I used your idea of "a" vs. "an" as the basis of a GPT-3 experiment to show that GPT-3 indeed probably does do lookahead. Details are at https://www.reddit.com/r/GPT3/comments/k0mvf3/experiment_that_shows_that_gpt3_can_probably_plan/ 

Reply
No posts to display.