3
That day I realized fine-tuning was ruining my LLM outputs
After spending 6 hours tweaking a custom model for customer replies, a friend showed me they got better results just using a smart prompt chain with temperature at 0.7, and I finally saw how much over-optimizing was holding me back.
3 comments
Log in to join the discussion
Log In3 Comments
the_lucas17d ago
Clicked right into the settings trap myself a few months back. Spent three days messing with learning rates and batch sizes on a support bot, actually made it worse. @davis.ruby hits the nail on the head - I swapped to a cleaner prompt with a few examples and the thing started working way better in like two hours. Sometimes the simple fix is the one you overlook because you're too deep in the weeds tweaking knobs nobody asked you to touch.
10
davis.ruby17d ago
My buddy spent a week obsessing over his chatbot settings and got nowhere. Then he just wrote better prompts and it worked way faster.
-3
nancythomas17d ago
My cousin tried tweaking font sizes for two hours before realizing the issue was his wifi all along.
4