ROUTEAgents & automation
Typesafe AI for Decision Making
Nav@NavDoesTech
Typesafe AI is a system for structured decision-making problems that selects from predefined answers and provides probabilities, rather than generating text like a traditional LLM.
Pattern↑ Jev AI decision models
What is typesafe AI?
alot of people are getting confused and thinking its an LLM like opus and fable.. but it cant write text at all. you define the possible answers in the request and it just picks one + gives you a probability for each. its basically for decision making structured problems!
Example: which model should handle this task?
LLM: "this looks like a refactor, id go with sonnet here since its cheaper and.."
Jev: (you give it the 3 options, it scores them)
sonnet: 78%
opus: 19%
haiku: 3%
your if-statement then just reads the 0.78 and uses sonnet
why is this interesting?
Because this allows us to swap out alot of the things we use LLMs for, for this much more efficient faster model and freeing up tons of compute. + alot of harnesses that do structured things can use this to make things more "efficient" like using this to route ur coding agent between models based on the task! similar to devin's fusion. oh yeah.. this model is AMAZING at games which have a set input(minecraft, pokemon and so on)