Hacker News
new
past
comments
ask
show
jobs
points
by
vadansky
13 hours ago
|
comments
by
jimmaswell
12 hours ago
|
[-]
Depends what you mean by "upstream". It would probably be inefficient to force the model to use more human-readable wording in its internal thinking traces, while translating a block of text at the end is a pretty trivial task for an LLM.
reply