Smart drunk interns: Punya Mishra on LLMs

A smart, drunk intern is Punya Mishra's description of LLMs in the "Wicked Problem" of AI and Why Algorithms Can't Replace Context — it made me laugh out loud. I've been describing the tools as clever, enthusiastic but inexperienced assistants — from now on, I'm going to include the word drunk!

While the article is situated within the context of education, it is a good read for anyone who uses AI tools in their professional or personal work. The exact quote from the article is: "smart drunk interns"; highly capable but fundamentally unreliable. Variable responses from AI are a feature of these models, rather than a flaw, writes Mishra, because they can generate multiple perspectives or examples. Of course, the user then needs to use their critical judgment to evaluate and filter the output.

💡 AI's effectiveness is still contingent upon informed human oversight, in the same way you would exercise judgment when reviewing the work of a smart intern, drunk or otherwise.