←back to thread

Everyone is capable of, and can benefit from, mathematical thinking

(www.quantamagazine.org)
669 points sonabinu | 3 comments | 21 Nov 24 01:45 UTC | HN request time: 0.002s | source
Show context
gsabo ◴[21 Nov 24 05:19 UTC] No.42201370[source]▶
I agree with the sentiment of this. I think our obsession with innate mathematical skill and genius is so detrimental to the growth mindset that you need to have in order to learn things.

I've been working a lot on my math skills lately (as an adult). A mindset I've had in the past is that "if it's hard, then that means you've hit your ceiling and you're wasting your time." But really, the opposite is true. If it's easy, then it means you already know this material, and you're wasting your time.

replies(18): >>42201667 #>>42201721 #>>42201729 #>>42201821 #>>42201915 #>>42202008 #>>42202194 #>>42202777 #>>42202780 #>>42202880 #>>42203292 #>>42203429 #>>42203649 #>>42204501 #>>42204671 #>>42206252 #>>42208029 #>>42219658 #
1. khafra ◴[21 Nov 24 08:16 UTC] No.42202194[source]▶
> If it's easy, then it means you already know this material, and you're wasting your time.

One thing I'm anticipating from LLM-based tutoring is an adaptive test that locates someone's frontier of knowledge, and plots an efficient route toward any capability goal through the required intermediate skills.

Trying to find the places where math starts getting difficult by skimming through textbooks takes too long; especially for those of us who were last in school decades ago.

replies(2): >>42204054 #>>42224086 #
2. llm_trw ◴[21 Nov 24 13:19 UTC] No.42204054[source]▶
>and plots an efficient route toward any capability goal through the required intermediate skills.

LLMs currently can't find efficient paths longer than 5 hops when given a simple itinerary. Expecting them to do anything but a tactical explanation of issues they have seen in training is extremely naive with something as high dimensional as math.

3. sn9 ◴[23 Nov 24 21:30 UTC] No.42224086[source]▶
LLMs hallucinate and are completely the wrong tool for this. People need to learn about the history of AI.

Expert-systems using techniques from the 80s are doing this right now [0].

[0] https://www.mathacademy.com/how-our-ai-works