Skip to content
Better HN
Top
Best
Ask
Show
New
Jobs
Search
⌘K
0 points
colejohnson66
5y ago
0 comments
Save
Share
Now I’m curious: what would happen if one
repeatedly
trained GPT-3 on text generated by a “previous generation” GPT-3? (Similar time successive JPEG saves)
0 comments
2 comments · 1 top-level
top
newest
oldest
AnimalMuppet
5y ago
· 1 in thread
I'm pretty sure it would drift further and further from something a human would recognize as intelligible.
datameta
5y ago
It would be a kind of DeepDream for NLP
j
/
k
navigate · click thread line to collapse