ChatGPT told me to fry lettuce

Here is what happened, as best as can be reconstructed. Someone sat down with ChatGPT, asked it something in the neighbourhood of cooking advice, and at some point in that exchange the model recommended frying lettuce. The story landed April 9th and has not moved since — one signal, no corroboration, no follow-up reporting. Now, to be fair to the machine: wok-fried lettuce is a real and genuinely excellent thing. Cantonese cooks have been tossing romaine over high heat for a long time, thirty seconds, a little garlic, some oyster sauce, the leaves collapse slightly and pick up a smokiness that bears almost no resemblance to a salad. The technique exists. The question the original report raises, though, is whether ChatGPT was doing something that sophisticated — or whether it was producing the kind of confident, plausible-sounding instruction that lands just far enough from reality to be useless or worse. That distinction matters, and right now the reporting does not settle it.

If confirmed as the latter — a genuine confabulation rather than an obscure-but-valid technique surfaced without context — here is what this means. The problem is not that the model got one recipe wrong. The problem is the confidence with which AI cooking assistants, nutrition advisors, and health-adjacent chatbots deliver instructions to people who have no particular reason to doubt them. Frying lettuce is low-stakes. The same structural failure — plausible syntax wrapped around unreliable content — applied to food safety temperatures, allergen substitutions, or medication interactions is not low-stakes at all. The people most likely to rely heavily on AI for this kind of guidance are also, frequently, the people with the least margin for error: new cooks, people managing restricted diets, anyone who simply cannot afford to throw the result in the bin and start again. One wilted lettuce is a comedy. The pattern it represents is not.

Watch for any follow-up that clarifies the exact prompt and full response — context here is everything, and if the model was actually attempting wok technique without labelling it as such, that is a meaningfully different failure than pure hallucination.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →