3 ms·
On a more-serious note, nah, I really don't feel threatened: LLMs are good at "throw high-probability results together from samples in training data", but not "
by Terr_ 3y ago
On a more-serious note, nah, I really don't feel threatened: LLMs are good at "throw high-probability results together from samples in training data", but not "logically analyze the final result to ensure it is sane", nor "change it to make it sane."
So improving LLMs may replace interns (or low-quality outsourcing) where an SWE has to review/fix everything anyway... but it'll take a fundamentally different kind of model to tackle those trickier essential tasks.