so much time and money lost due to llms setting awful defaults somewhere in code and it being based on pre-training knowledge
"oh, you want the runs to be fast? max_tokens=128 should be enough, then"
10:28 PM · Jun 3, 2026 · 4.7K Views
Prime Intellect's Florian Brand says LLMs frequently generate code with suboptimal default parameters like max_tokens set to 128
These pre-training patterns waste developer time and cause execution failures.
--0--
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
VIEWS308LIKES11
xlr8harder@xlr8harder
@xeophon oh god max_tokens is such a frustrating one
Florian Brand@xeophon
so much time and money lost due to llms setting awful defaults somewhere in code and it being based on pre-training knowledge
"oh, you want the runs to be fast? max_tokens=128 should be enough, then"
7hViews 308Likes 11Bookmarks 0