Not “train a bigger model.”
I mean something you genuinely want to know but nobody sane is spending the compute on.
Weird architecture, absurd context length, thousands of agents, ridiculous ablation, whatever.
What are you testing?
Guillem
Posted on
If you had unlimited compute for one experiment, what are you running?
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (1)
Phising for ideas? :D