#ai #llm
# [[Epistemic status]]
#shower-thought #to-digest
# Related
> [!TODO] Related
# TODO
> [!TODO] TODO
# LLM fine-tuning is obsolete
Too slow process. Need to deal with many models etc.
You rather do prompt engineering with semantic search instead
## Few experiments
- Using text-davinci-002 with 5 shots makes it 4 times slower than a fine-tuned model
- base models are 5-10 times cheaper
- #metaphysical but fine-tuned models reduce creativity - kind of make sense statistically?
# External links