ChatGPT strikes again

We stumbled across a report recently about a youtuber who posted a story on how ChatGPT nearly killed her best friend. According to the story the friend used Chat to try help her identify a plant in her garden, and it did not go well. The friend sent Chat a photo of the unidentified plant, asking 'what plant is this,' only to be told it looks like carrot foliage. According to screenshots, ChatGPT went on to list several reasons it was 'confident' the plant was carrot foliage, including the 'finely divided and feathery leaves,' which is very 'classic' for carrot tops. Interestingly, the chatbot went on to list some common lookalikes of carrot foliage, including parsley, cilantro, Queen Anne's lace, and poison hemlock.

The plant was in fact hemlock, and the user was met with multiple assurances from Chat that it was not poison hemlock when she queried it. The friend was also encouraged by Chat to incorrectly label the plant as carrot foliage on the assumption it might be in a shared garden in the school where she works. 

Thankfully she wasn’t silly enough to take her advice solely from Chat and did some more digging to identify the hemlock correctly.