#ThisWeekinAI I learned that LLMs can’t count.

Like, they can summarize 1,000 rows of open text survey feedback but they can’t tell you that they analyzed 1,000 rows. Worse, they’ll tell you “128 rows” or some other nonsense.
Similarly, while LLMs can be very good at augmenting hashtag#SQL writing, at least through helping with pesky syntax, they don’t know they’re wildly off in their very official looking responses. SQL is pretty hard to debug, at least for me, and so being sent on a wild LLM goose chase can be an especially malevolent time burner.
Bottom line: Surprise! LLMs are sociopaths who lie, seemingly, for the joy of it. Like, instead of ‘Claude’ or ‘Gemini’ or ‘Khanmigo’ they should all be called Ripley. As in, “Talented Mr” not Sigourney Weaver from Alien.

I don’t think it is an accident that AI has hit education the hardest. That is, the most change caused by AI has been IN education. Sure, companies are talking a big game – cost reductions etc. But most corporate folks are too busy to investigate beyond perhaps throwing together a chatbot as a better search frontend or doing some light brainstorming. Students, though? Especially the ones struggling, who have little critical thinking skills to begin with and can’t discern quality from dreck? Who are operating in environments with little rigor (generic 5-paragraph essay without typos earns a B+)? They’re all in…and upending the apple cart.