Prompting with Caution: AI, Mushrooms, and the Art of Not Getting Poisoned
The image below, currently circulating around AI conversations, is fun and strikes a very clear point on how AI prompts are both misunderstood and misused, leading to such problems as “eating poison mushrooms”.
Prompts are not a one-and-done Q&A, and how prompts are started determine if the desired answer will be achieved. In this example, a better way to start (and there are other, stronger starter prompts, I am sure) would be to ask the Gen AI:
Are you able to identify this plant from the picture? Please provide at least five possibilities of what this plant’s name might be.
The AI should give you its best guess, and from there you need to determine if (a) the AI is in the ballpark of being right or (b) if the AI is so incredibly off-base that you either need to re-engineer your prompt, or find another Gen AI tool to help, or find a botanist.
Let us assume a botanist is not readily available but you do have a bag of Gen AI resources to continue to help you, and you are satisfied that your Gen AI probably has made a correct choice. Your next prompt may be:
Of your possible plant identifications, are any of them poisonous when consumed by humans? Please provide source references.
The last point is key – don’t just take the Gen AI’s word for it, verify as best as you can with corroborating data. …and hopefully you get the idea, that you could go on in this fashion for several more engagement prompts, until based on your knowledge, experience, and risk factors, if you truly want to attempt to ingest that mushroom!
