Boosted by brib@bribstodon.xyz ("brib :neofox_floof: :Nonbinary:"):
dalias@hachyderm.io ("Cassandrich") wrote:
@mos_8502 @mhoye Regardless of size a language model is just an extraordinarily bad way to produce code. For that you want rigorous processes not vibes.
These models are mostly bad for generative use anyway, but small, specific, curated ones could be useful for helping the user find patterns. In a code context, patterns of common bugs and vulnerabilities, idioms used, seeking out existing implementations of an algorithm to read and learn from or to select as a library to use and credit, rather than to automate plagiarism of somebody else's implementation, etc.