Perhaps somewhat tangental, but your breakdown of how LLMs use code to bypass their math deficiency reminded me of the concept of Core Math. As a non-parent, I had never heard of this before, but apparently school teach math a completely different way than it used to when I was a kid.
In this new way, math is taught by breaking problems down into easier steps instead of 'remembering' long formulas.
In a roundabout way, this might prepare today's kids to better understand how AI tools like chatgpt etc. handle the math questions in the middle of pure LLM back & forth, as that's pretty much exactly what they do. You present a problem (in human language), LLM understands it, then breaks the math down into actionable chunks and creates simple code to solve it step by step, then responds using LLM.
Quite fun that kids & LLMs are learning the same approach, even though that wasn't really planned!
Thank you Alex! My kids are now teens and I had to understand Core Math to help them with their homework - embarrassing since they were convinced I didn't know basic math, :)
From what I understand, Core Math was a reaction to computers - kids are now surrounded by computing devices that make arithmetic trivial. The important skill is not completing the arithmetic to absolute precision, but understanding how to quickly think numerically and hone in on a good answer.
To your point about LLMs, I think we'll need a new approach like Core Math, but for critical thinking. LLMs will make language generation trivial, so the really important skill is understanding how to do the higher order understanding, reasoning, and creating that can guide the LLM. I believe (and I hope!) there will always be some kind of thinking that is outside the capability of AI (even as we approach AGI), and that our kids will benefit from learning how to learn on their relative strengths and where to pull AI into the process.
Great post, Ravi.
Perhaps somewhat tangental, but your breakdown of how LLMs use code to bypass their math deficiency reminded me of the concept of Core Math. As a non-parent, I had never heard of this before, but apparently school teach math a completely different way than it used to when I was a kid.
In this new way, math is taught by breaking problems down into easier steps instead of 'remembering' long formulas.
In a roundabout way, this might prepare today's kids to better understand how AI tools like chatgpt etc. handle the math questions in the middle of pure LLM back & forth, as that's pretty much exactly what they do. You present a problem (in human language), LLM understands it, then breaks the math down into actionable chunks and creates simple code to solve it step by step, then responds using LLM.
Quite fun that kids & LLMs are learning the same approach, even though that wasn't really planned!
Thank you Alex! My kids are now teens and I had to understand Core Math to help them with their homework - embarrassing since they were convinced I didn't know basic math, :)
From what I understand, Core Math was a reaction to computers - kids are now surrounded by computing devices that make arithmetic trivial. The important skill is not completing the arithmetic to absolute precision, but understanding how to quickly think numerically and hone in on a good answer.
To your point about LLMs, I think we'll need a new approach like Core Math, but for critical thinking. LLMs will make language generation trivial, so the really important skill is understanding how to do the higher order understanding, reasoning, and creating that can guide the LLM. I believe (and I hope!) there will always be some kind of thinking that is outside the capability of AI (even as we approach AGI), and that our kids will benefit from learning how to learn on their relative strengths and where to pull AI into the process.