does claude actually make stuff up or is that just chatgpt
writing blog posts with AI and keep hearing about hallucinations. does claude sonnet 5 just invent facts or is it reliable enough for content that goes on a real site? iirc someone said claude is better about this than chatgpt but could be wrong
all of them do it. frequency depends on the domain - obscure stuff way more than common knowledge. maybe 1 in 50 responses on mainstream topics, 1 in 10 on niche library docs
ok so 1 in 50 on mainstream topics is actually way higher than i thought. this is kind of a problem for anything customer-facing where you cant manually check every output
all of them do it. i stopped trusting any fact it gives me unless i can check it myself. fine for drafts, not for anything with real numbers going out with my name on it
they all make stuff up. claude hallucinates less than chatgpt in my testing but it still happens, especially on obscure library versions
Can you give me a percentage or frequency? Like does it hallucinate 1 in 10 responses or 1 in 100? I need to know if this is safe enough to use for customer-facing content or if I'm going to spend more time fact-checking than it saves me.
so is it safe enough for work stuff or should I just write everything myself. like if I use it for report and it makes up statistic, this is bad for me