The internet is full of false recipes that can't possibly work, probably increasingly LLM generated themselves. Consider: Persimmon jam. Not possible short of some secret chemical process-- heating reverses the chemical reaction that eliminates the astringency of the unripe persimmon. Yet there are now countless recipes online that will lead you to inedible muck, all effectively bogus hallucinations. Not necessarily machine ones, but also just content farms manually making every conceivable permutation of every existing recipe, the author of those persimmon jam recipes has probably never seen the fruit first hand.
At least if you LLM them on your own you know to be dubious and also you're more likely to use a 70B model (or GPT4) which is a little more likely to get them right, vs whatever garbage the spammer used.
It would probably be pretty interesting for someone to make a concerned effort to digitize every recipe in print prior to 2005 or so-- the relevant threshold time isn't when LLMs came on the scene but when google made low effort content farming so profitable.
It wasn't guaranteed that every recipe printed pre-internet had actually been cooked by anyone, but it was a lot more likely.