I started using ChatGPT, Copilot, and Meta AI to generate random sets and cartesian products and do stuff with them.
Sounds interesting. I only know that excel is a bad rng. Has there been any discussion/research on how good ai is at such a thing?
During my divorce, I asked ChatGPT a few basic legal questions, which saved me a little bit in legal fees, which was nice. One time I wrote some SQL code using logic not supported my the SQL engine I was using. The AI assistant rewrote my code so it would work, which was appreciated.
1 Like
Went to see Sal Khan yesterday, talking about Khanmigo and their partnership with Microsoft and the future of AI. Mostly focused on education and some of the tools they are developing to help teachers integrate AI into their curriculum in a positive way.
Interesting to see his perspective on it, and gives me some hope that we can actually leverage AI to make some positive changes to the way students learn.
2 Likes
ChatGPT can access the internet now…so, I asked it on November 01…
Well, it’s intelligence is artificial. 
Except it is not an infinitely patient genius. It is a model fit to a large codex of human discourse. Humans who, from time to time, suck. And are astoundingly bad at recognizing the truth.
Which is why you also end up with stories like
But sure, let your kid be taught by a system pushed by the same people who built social media “engagement” algorithms. I’m sure it will be fine.
I suspect this was a prank played by the user, using invisible characters or some other prompt injection trick.
So it’s an infinitely patient genius which can slip into death threats, profanity, or racism if you say the wrong thing to it. Good thing children never say anything odd that could throw it for a loop.
I don’t mean “odd”, I mean someone intentionally tricked it by saying “repeat after me…” and then making those instructions invisible or putting them in a file or picture.
If you mean to say that children will hack or otherwise abuse LLMs, yes, I agree. At the moment I’m more worried about them using AI to create nudes.
Agree that it is too soon for children to use the current versions.
They are still too unstable.
I have not (yet) unleashed it on my own kid.
As a tutor, I don’t think it would help my daughter much right now. She can just ask me the same questions. Or read books on her own.
As a teacher, we could just use Khan Academy, which is far superior to her own teachers (who aren’t teaching her anything). The problem is I don’t want her staring at a screen alone more than she already does.
That said, it would be a really marvelous toy. She always wants to play pretend with me. More than anything, and I always lose my patience eventually. It could even add illustrations, and voices and videos. It would be the best toy ever. I shy away at the thought because it’s too good in a way.
Same with image generation. I want her to feel a sense of excitement and pride around the art she creates… which I think might be lost if she gets in the habit of asking a computer to create it for her.
Of course, that’s just me and mine, speaking from a place of privilege. Not every dad in the entire world has the time and resources to answer every possible question in every language.
And to this point, all it does is predict future words. It has no understanding of truth vs fiction, an external world, values, etc.
Instead, it’s just a big joint probability distribution of how words appear together.
1 Like
Well, this should be interesting…
1 Like
