causal
Public experimentation amplifies hype via cherry-picking
GPT-3's public website let people experiment, and because it succeeds only about 4 times in 10, users cherry-pick the good outputs and post them on Twitter, creating tremendous hype.
causalpending
Speaker
DanielEvidence Quote
“The results it produces, I would say, maybe 6 times out of 10, are bad.”
Created: 11/13/2026, 12:47:43 AM
Your notes
Loading notes...