[HN Gopher] Synthesis Tutor - Your child's own superhuman math t...
___________________________________________________________________
Synthesis Tutor - Your child's own superhuman math tutor
Author : brennancolberg
Score : 64 points
Date : 2023-06-08 18:36 UTC (4 hours ago)
(HTM) web link (www.synthesis.com)
(TXT) w3m dump (www.synthesis.com)
| pimlottc wrote:
| If you're pushing a personal education tool that uses AI, you'd
| better explain exactly how AI is used before I'm going to put it
| in front of a child.
| ClintFix wrote:
| I had the opportunity to work on this. I let my son try the demo
| last week. Hours afterward at dinner I asked him to explain
| binary to my wife.
|
| "Mom, imagine you have the number 101 in binary. The first 1 is
| the fours spot, so you have 4. There's none in the twos spot. and
| 1 in the ones spot. So that means you have 5."
|
| Two days later he came into my office and asked if he could do
| more tutor.
|
| That's a huge win in my book!
| symic wrote:
| I've taught math at the college level for 25 years. The video
| looked impressive. Do you know up to what level of math it
| does?
| rahimnathwani wrote:
| If you're looking for stuff at high school or college level,
| check out mathacademy.com
|
| (No affiliation - just a happy customer.)
| slackerIII wrote:
| How does that compare to Khan Academy?
| rahimnathwani wrote:
| How it's similar:
|
| - focused on mastery
|
| - good treatment of math
|
| How it's different:
|
| - doesn't cover elementary school topics
|
| - lessons are text, not video
|
| - paid, not free
|
| - lessons are presented based on what you already know
| (e.g. you can't choose to study a topic before you've
| demonstrated mastery of all topics it depends on)
|
| - I pick the next lesson from a very small set (usually 5
| options, sometimes fewer), so there's no wasting time
| choosing what to study next
|
| - really great spaced repetition system that prevents
| forgetting
| symic wrote:
| Thanks for the head's up.
| poorbutdebtfree wrote:
| $50/month is pretty steep for what looks like a Khan
| clone.
| 2snakes wrote:
| It is accredited though. College credit?
| avhon1 wrote:
| Further down the page, they have a "Curriculum roadmap",
| which goes up to things like "College Algebra" and "Matrix
| Theory".
| ClintFix wrote:
| We plan on having math through age 10 by the end of the year
| vintermann wrote:
| Just like that language self-study tool that was on the front
| page recently, an ML based math tutor is one of the things that
| are much easier to imagine and to hype than to actually make work
| well.
|
| I just don't trust companies that try to give the impression that
| they've worked it out already - and this one seems even worse at
| that than the language thing.
| yding wrote:
| Cool! Would love to beta test with my 10 year old.
| gnicholas wrote:
| How does this compare to Khanmigo, Khan Academy's AI chatbot?
| ClintFix wrote:
| It has manipulatives/widgets that the tutor works with and
| responds to, rather than just a chat bot.
| light_hue_1 wrote:
| I tried the demo.
|
| As an educator who has taught math: this is frustrating junk. It
| moves at a crawl. It doesn't explain the reasoning for anything.
| It provides no motivation.
|
| It's full of incorrect statements like "Computers struggle with
| too many digits."
|
| Pretty much the worst practice if you want to teach someone
| anything.
| symic wrote:
| In third grade parlance it is correct to say computers struggle
| with too many digits.
| mywittyname wrote:
| They don't "struggle". Computers exhibit definite behavior
| when a program attempts to store or operate on numbers that
| are too large for the allocated space.
|
| For integers, overflow behavior is pretty easy to understand.
| symic wrote:
| Obviously I know this. As stated, in 3rd grade parlance
| it's appropriate to say "they struggle". Have you ever
| heard someone say, "I need a new computer; mine's dying."?
| This level of pedantic nitpicking is not appropriate for
| 3rd graders. I've heard professional programmers say that
| their computer struggles with compiling certain programs
| they are working on. Colloquialisms can be useful and they
| are ubiquitous.
| burnished wrote:
| Not the allocated space, the bit width of the ALU. Besides,
| your explanation doesn't add anything to the description of
| 'struggle' beyond 'in a normal and expected fashion'.
| musicale wrote:
| Not sure what "struggle" or "too many" is supposed to mean,
| but my laptop (running Python in a Linux VM) works fine for
| ~5m digits: >>> import math >>>
| print(math.log10(math.factorial(1000000)))
| 5565708.917186718
| symic wrote:
| A computer can't display more digits than the number of
| particles in the observable universe. This is a finite
| number. As such it is correct to say that computers
| struggle if the number of digits is too large. While for
| practical uses modern computers don't struggle with finite
| integers one normally encounters they do struggle with
| precision in certain circumstances. For 3rd graders I think
| it's OK to introduce them to the concept that computers,
| like all devices, have limitations.
| avereveard wrote:
| What a wild generalisation. I think math is a special case, but
| history and literature at a scholastic grade should be fine
| ilovecurl wrote:
| "Frustrating" is putting it mildly.
| local_crmdgeon wrote:
| I think the impact of LLMs for education is REALLY
| underappreciated. You don't need AGI, just local repetition, with
| some interactivity. Your average science teacher knows a lot
| about a lot, but not a ton about anything. That's perfect for
| one-on-one LLM tutoring
| acomjean wrote:
| I get that people want this.
|
| The issue is that these things go off the rails sometimes.
| Without any supervision and as an "educational" resource it
| will be unquestioned and in front of people who don't know
| enough to question it.
| burnished wrote:
| Eh, its a lot easier to question a machine that isn't also
| defending its ego. Kids are quick to spot and question a
| contradiction until that behavior is socialized away - an LLM
| doesnt get impatient when you ask a bunch of questions, or
| dismiss you because it thinks the question is dumb.
|
| It can be a good tool.
| elevaet wrote:
| I think we have to just really drill into our minds that, as
| convincing as these tools are, sometimes they are flat out
| lying/mistaken/misinformed. If we treat them like our know-
| it-all friend who is sometimes just a big bullshitter, and
| take everything with a pinch of salt, then these can be
| extremely useful tools for learning the big picture if not
| the specifics.
|
| Now, is it reasonable to expect children to be skeptical
| about everything the AI tutor is telling them? Probably not?
| Maybe?
|
| We've all had human teachers make mistakes or bullshit, maybe
| we can accept some fallibility in these LLMs too.
| local_crmdgeon wrote:
| Exactly - people want these to be infallible. I say stupid
| stuff literally constantly and I'm a grown adult. Teachers
| are incredibly fallible - but someone schooled in math can
| still do a ton of good.
| lumb63 wrote:
| More generally, we should be skeptical of all new
| information we learn. What makes someone more trustworthy
| than an LLM? For science we need experiments, theories,
| mechanisms, etc. For history we need sources, ideally
| primary, and working from there. For math we need proofs.
|
| Knowledge is much more fallible than many appreciate. Many
| of the "universal" truths of today were once unknown or
| controversial. Our understanding of the world is built on a
| giant set of shoulders, of all those who have come before
| us. We do ourselves a great disservice not to acknowledge,
| understand, and appreciate _how_ something came to be
| "true".
| acomjean wrote:
| > What makes someone more trustworthy than an LLM?
|
| I mean people can be terrible. But a teacher is a trusted
| profession and likely needs that paycheck and will avoid
| doing stupid things to loose that.
|
| I don't think LLM care one way or another. Someone should
| ask one.
| staunton wrote:
| > Now, is it reasonable to expect children to be skeptical
| about everything the AI tutor is telling them? Probably
| not? Maybe?
|
| That might be the single most important thing to teach to
| children now.
| da39a3ee wrote:
| This is madness. I'm trying the demo and it is completely
| incomprehensible to me. This would have been incredibly
| frustrating to me as a child -- I would want them to stop
| speaking in metaphors and tell me what the fuck they were trying
| to teach me. Since then I have acquired a scientific PhD and
| taught myself undergrad computer science and math, and I do not
| understand this lesson.
| ClintFix wrote:
| Weird because my 8yo understood it and loved it
| zinclozenge wrote:
| I can't wait until this tech progresses to the point it can
| reliably tutor for more advanced courses like real analysis,
| group theory, linear algebra, etc. I actually wondered if it
| would be feasible to use NLP techniques to check user submitted
| answers to proof based problem, LLMs might be get there.
| sytelus wrote:
| I don't see anything functional here. Is waiting list
| announcement all it takes to get on front page these days?
| gnicholas wrote:
| There's a short demo you can try. It doesn't work on mobile
| though.
___________________________________________________________________
(page generated 2023-06-08 23:00 UTC)