Teachers at Bloomington Kennedy High School are taking a practical view on artificial intelligence, teaching students how to use the technology and helping them understand where it can spin out of control, a sort of driver’s ed approach to AI.
Schools teaching “how to use AI” are missing the point: these things aren’t magic machines, they’re guessing games.
LLMs have no ability to reason or understand anything. They just regurgitate words using a probabilistic model. They are little more than a fancy autocomplete, and autocomplete isn’t always right (see also DYAC).
Even as statistical parrots and all that, they’re fairly useful for learning, you just have to know their limitations. They’re very useful for questions that are hard to search for but easy to verify, like “What is this funny math symbol?”. Especially when the symbol has different meanings in different contexts, and you can explain the context to the LLM.
They are not useful for learning if the information they give is wrong.
Neural networks are a tool and they have their uses, but something that generates garbage that is factually wrong a significant percentage of the time while burning our world to do it is not the tool our students need.
Schools teaching “how to use AI” are missing the point: these things aren’t magic machines, they’re guessing games.
LLMs have no ability to reason or understand anything. They just regurgitate words using a probabilistic model. They are little more than a fancy autocomplete, and autocomplete isn’t always right (see also DYAC).
https://garymarcus.substack.com/p/llms-dont-do-formal-reasoning-and
Heh, you can see my opinion on Gary Marcus in another comment of mine: https://midwest.social/post/17934593/12838629
Even as statistical parrots and all that, they’re fairly useful for learning, you just have to know their limitations. They’re very useful for questions that are hard to search for but easy to verify, like “What is this funny math symbol?”. Especially when the symbol has different meanings in different contexts, and you can explain the context to the LLM.
They are not useful for learning if the information they give is wrong.
Neural networks are a tool and they have their uses, but something that generates garbage that is factually wrong a significant percentage of the time while burning our world to do it is not the tool our students need.