💭 baffling that absolutely everyone technical or otherwise seems to think of llms as chatbots at their core rather than completion engines. At some point I expect some revolutionary finding 'guys you can drop the chat framework and get better results on some problems' as if there's any reason to think otherwise.