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?
Back to Articles
If you had unlimited compute for one experiment, what are you running?
Not “train a bigger model.” I mean something you genuinely want to know but nobody sane is spending...
Author
Guillem
@uguillemStats
Reactions0
Comments1
Read Time1 min
Published
You Might Also Like
Harsh Aug 10, 2026 • 7 min read
You Don't Have an AI Problem You Have a Thinking Problem.
I used to think AI was making me lazy. I was wrong. AI wasn't making me lazy I was using AI as an...
#ai#programming#softwaredevelopment#productivity
56 37
Roberto B.Aug 12, 2026 • 7 min read
The Next Evolution of Software Developers
The next evolution of software developers: from implementation to intent, orchestration, and...
#ai#learning#mentorship#software
45 17
