Just wondering since the knowledge cutoff for LLMs is about a year or so behind the current date.
It would be nice if the models were able to know how to use the type annotation syntax when it comes out, instead of having to wait until they are properly trained on enough material to be able to use it.
I’m sure there are ways to seed their knowledge when they have to use it, but I feel like the sooner the info is made public, the better.
1 Like
Short answer is no. I cover our current status in my recent ElixirConf EU keynote. Feel free to skip to the end of the video: https://www.youtube.com/watch?v=Ay-gnCqDw9o (starting from the Elixir v1.20 slide).
5 Likes
Thanks for the reply. Obviously it will be ready when it’s ready, and we can use a skill or some kind of prompt to seed the typing syntax into the context in the interim if everything comes too quickly. Just had to satisfy the burning curiosity!
I was thinking of starting to use Typle once Elixir 1.20 lands to that same effect that you described.
1 Like
I’d pay money to become a fly on the wall listening to you and Chris talk about LLMs. But not too much money because I assume for the procedure to turn myself into a fly and back would already be pretty costly.
4 Likes
New revenue stream for the type system development? 
1 Like
I’m trying to build AOT compilation for BEAM bytecode. It sounds interesting if we can bring this to a lower level to reduce type checking.