@adrianco@mastodon.social @norootcause@hachyderm.io fwiw, I talked to a guy who has been coding with LLMs since gpt 3.5 Context is one thing, but attention is another. Using prompts that do monad lifting give substantial gains in functional languages. When state is in external mutable variables, it requires attention which is separate from context. Using monads, the state is isolated and the LLMs can just “emit the next token” Also, he’s a self-described “C/C++ guy” not a scala or Haskell freak