It’s about the quality over quantity. I’d bet there’s significantly more bad React published online than Elixir. LLMs don’t know what’s bad and what’s not, only what’s more likely. How much bad Elixir is published online?
I’d also bet the nature of the language itself would only affect how few samples it can train on before appearing competent
I’d also bet the nature of the language itself would only affect how few samples it can train on before appearing competent