Mushroom hunting with LLMs: what can go wrong?
A journalist examined the risks of using large language models (LLMs) to identify mushrooms in the wild. While LLMs like Gemini 3.6 and 3.7 Flash performed well at visual tasks, they had concerning error rates when distinguishing edible from poisonous mushrooms. The top models made a 12% false negative rate and a 12% false positive rate, with one deadly mushroom misclassified as edible in their results.
Experts emphasized that photographs alone were insufficient to reliably identify mushrooms, highlighting the need for expert verification and DNA sequencing. The author cautioned against trusting AI to determine mushroom safety, warning that such errors could have fatal consequences, especially for inexperienced foragers.
Written by urgent.news from Hacker News's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
