Reaction
The compute trade-off between web search and AI models
A post quotes @paraga arguing that spending more compute on web search can save compute on the model, but the balance depends on how costly the model is to run.
TLDR
A post shares @paraga’s view that search and model compute are an economic trade-off. They suggest doing less search work for a cheap model such as Luna and letting a bit more information into its context. For Fable, which they describe as expensive in time and money, they favor doing more work before passing information to the model.
Combined views
39.8K
3 Sources, first seen ago
96 likes31 comments29 saves12 reposts
