Thanks very much! Too bad about Genie :( Curious what about khanmigo felt better than just using chatGPT? I heard they had some problems with factual errors, but maybe have improved since?
Disclosure: I work in risk management, and a component of my work has been providing subject matter expertise in implementing generative AI with controls and guardrails in large enterprises. Thoughts and opinions always my own. I pay for Khanmigo (but also ChatGPT, Claude, Kagi, etc).
TLDR I'm more confident in Khanmigo's safety and security (tailored to a young, non adult audience) of the product than I would be of OpenAI and ChatGPT, based on my research and discussions with others in industry. Khan Academy incentives are not OpenAI incentives. I trust my kids alone with Khanmigo, I would not trust them alone with ChatGPT.
Empire of AI By Karen Hao [1] is a great read by the way, and helpful context adjacent to this topic.
Ask any public school teacher and they will tell you that the number one determinant of success for a student comes down to parental involvement. They might also tell you that school administrations ignore and sometimes actively fight against parents getting involved. Homeschooling just takes parental involvement to the extreme. Public school has a lot more to do with keeping kids off the street and giving parents the hours to work -- day care -- than with education. Some kids get lucky and get a lot out of school, but a lot more don't.
All these factors will contribute.
Explosion of tech resources to help caregivers teach kids, most notably, but not limited to chatGPT
Decline in federal funding for schools
Growth in homeschool communities and entrepreneurship by teachers leaving the system (as they develop classes and curriculum in their communities)
Clear benefits of 1-1 mastery learning
Clear benefits to mixed age groups with more time to interact
Clear benefits of more time for self-directed learning
Clear benefits of choosing different curricula and learning plans to suit different children's learnign needs
Pushback on teaching climate change and inclusive curriculum (Black history, LGBTQ+, Indigenous history, etc) leading to mass exist on the left
Rise in government funding for modular learning (ESAs and microgrants)
Rise in remote work (families can travel and be at home with kids; when kids start young they can be independent while parents work)
They use it at her school, and I've bought her some books with projects for her because she's liked to do things outside of school too (for example, ML for kids, which is odd to think about but that's what it is). We've worked through some of the projects in the book, she's made simple pong-style games, and some other things.
Her school has kind of gradually been working through apps and scratch to build up concepts. Some of the early ones were games where you had to program a route out of a maze or something like that by doing things like providing the steps (up down right left, number of each) out. That kind of led to other things, which led to other things.
It sort of built from "tell this robot how to get out of a maze" to "tell your ipad how to do X" where X got gradually more complex over the course of months or a couple of years.
This might be the wrong website upon which to poo-poo LLMs, but as a counter-observation: the trick with LLMs is knowing how to evaluate their output for correctness, since hallucination is a very real risk. Professionally, reading forums, blogs, and the Stack Exchange network carries similar risk, but also comes with (a) at least they were trying to be helpful (b) the voting and moderation system as a mild form of reining it in
Thus, unless one wanted to make a "ParentalLLM" through which one could submit questions to an ourhouse.internal/api/v1/prompt api and get back "chat" responses over sms/some existing chat protocol, I'd carefully consider the risk verses reward of bringing LLMs into that process
I also recognize that I came up in the era of autodidactic lessons, and that predisposes me to "teach yourself how to fish" style approaches, but just wanted to draw attention to those LLM interactions coming with trade-offs
I don't know what the right web-site would be to "poo poo on LLMs" :) It's here or nowhere, I think. Also, thanks so much for these nuanced reflections on the pedagogy. Super helpful.
