Mushroom hunting with LLMs: what can go wrong?
Summary
The article explores using LLMs and vision models to identify mushrooms from photos, examining a large Danish dataset and a benchmarking study of multiple models. It highlights common mistakes, safety concerns when relying on AI for health-critical decisions, and emphasizes that human verification and multiple photos are essential; Gemini 3.7 Flash performs well on accuracy and cost, but AI should not be trusted for safety decisions.