GPT-4 still confidently makes up sources for wrong answers and throws subtle mistakes (the obvious mistakes aren't as big a nuisance) into output. This isn't t

GPT-4 still confidently makes up sources for wrong answers and throws subtle mistakes (the obvious mistakes aren't as big a nuisance) into output. This isn't to say gpt-4 isn't cool or impressive or a development to watch and learn about and be excited about, but I frequently see criticism dismissed as "you must be using 3.5" while I find 4 still costs more time than it would have potentially saved.