My little sister's use of ChatGPT for homework is heartbreaking

>If you'd told people 50 years ago that "literacy" would decline to the level of the reddit poster's style of writing, with poor sentence structure and words in ALL CAPS, they'd be pretty shocked. Now it's just normal communication.

I would argue that the people who would have written like 50 years ago just never had the opportunity or audience that people do today. Except maybe in a Letter to the Editor of a newspaper or something.

So many more people are expressing themselves in public than ever were able to 50 years ago so I think it just seems that people have become less literate than they were. So many YouTube comments and Facebook posts that never would have occurred 50 years ago.

It's a democratization of the broadcasting of expression more than literacy decline.

> Tests/quizzes are good teaching tools.

I skimmed the linked material and that's not my conclusion. Practice tests are better teaching tools than just memorizing the material and then taking a real test, but on an absolute scale, I'm convinced that they're pretty terrible for most types of material.

Throughout my education I've come across countless people who could memorize the material and recite it with relative ease but they didn't have any intuitive understanding so they couldn't use that "knowledge" to solve real problems. Much like ChatGPT itself.

Equating good memorization and recall performance with good education and knowledge seems like a form of cargo culting to me, it's missing the essence of what makes knowledge powerful in the first place.

We're in Canada; doing the same. But most of my kids' classmates are on YouTube all the time. The brainrot content is unavoidable and it's creating issues for my kids' socialization.

I believe uploading neural weights in the form factor of a reaction speed game will become the norm.

Imagine the token weights etc to live an an N dimensional embedding.

Consider a random 2D isometric projection, now imagine plotting (a subset of) the tokens on screen at their respective locations, with the background having patterns conveying the orientation of the projection.

Imagine tokens appearing FIFO, and 95% or so of the tokens at their correct position, and 5% at incorrect positions.

The user is expected to identify misplaced tokens in the projection.

Each the frame the perspective slowly changes.

Since we guarantee ~95% of the tokens on screen are correctly placed this enables the human to absorb their correct high dimensional location subconsciously.

Imagine this game is made a bit addictive.

Imagine eventually the user gets high scores and thus knows most of the coordinates of the token embeddings in a coordinate free way (no axes had to be drawn)

At that point all the information the ANN required is undisputedly present in the humans brain, absent how those coefficients are used. This is where the fun starts.

The alcorithm can calculate the likelihood of observing 2 tokens in a certain order. It can generate more probable pairs, this will correlate with the information in the brain.

For example imagine the user knows some spanish, and that the language model was multilingual. Suddenly the brain starts picking up the correlation between close juxtapositions of certain tokens, and their positions learnt from the computer game.

While playing and getting better scores in the game, the user starts noticing its own grammar and vocabulary improve, because the brain has been helped in better estimating next tokens...

People will be able to learn math, languages, programmnig languages, tables of chemistry, etc... with substantially less effort.

> What is stopping you from whitelisting websites she can visit?

You're right. I have an anecdote though. My kid is a good student, does his homework, and does well on tests. He has a time-limited device he can use at home with a few whitelisted websites. But one of the services we allow him to use is Perplexity. He's a naturally inquisitive person and this is how he learns, so an AI is a perfect fit for when his parents don't have the answer to his question, or if he wants to ask detailed follow-ups. He's young enough that he doesn't have much homework (and we wouldn't let him use his device for it anyway), so cheating isn't as much of a concern yet. We do discuss it though, and will keep that conversation going as he ages.

This won't be solved by abstaining from AI entirely because a) kids will not always be under our control, so they need to learn how to self-regulate, and b) knowing how to properly (and ethically) harness AI is going to be more and more important to their future success.