Post #2672628

@nuthatch@infosec.exchange @jackscottau@aus.social @circafuturum@infosec.exchange LLMs can only be as useful as the average quality of relevant content contained in their training set. As (1) more and more public code becomes enshittified by LLM contributions, and (2) fewer sites like Stack Overflow remain to churn out high quality solutions to common problems, then the training data quality will deteriorate and so too will the code generated by LLMs.