Hacker News
new
past
comments
ask
show
jobs
points
by
p-e-w
49 minutes ago
|
comments
by
bavell
33 minutes ago
|
[-]
Instead of pre-training, I'd say it's more like a better architecture, training regime, higher fidelity data, and perhaps better initial "weights".
reply
by
LoganDark
26 minutes ago
|
parent
|
[-]
All of that stuff was still trained by evolution, though.
reply