Dunno, we pretty much know our current classroom education system is abysmal compared to tutoring[1], which is on the other hand prohibitively expensive. Managing to bridge that gap would be an enormous benefit to humanity.
For kids that are curious and want to learn - access to something like gpt-4 is amazing. All I had were guides like Eric S. Raymond's how to be a hacker that I had stumbled onto while trying to learn. It was so much harder to learn programming back then if you didn't know where to start and nobody around you knew about it.
With gpt-4 you can just ask and get examples.
My little brother is doing some form of machine learning in grad school currently and when I saw him last around when ChatGPT was starting to become a mainstream talking point we discussed a lot about how immediately going to something that can answer a question (e.g. how do i write code that does X) is tempting but, can be a bit dangerous when we offload our own ability to reason through something to an external entity.
Shouldn't we be fostering thinking patterns that allow for creativity and innovation when given very little, versus fostering a sense of "oh I can always get the right answer to a question whenever I need it. So why bother with building a general mental problem solving framework for myself"?
https://en.wikipedia.org/wiki/Is_Google_Making_Us_Stupid%3F
https://www.theatlantic.com/magazine/archive/2008/07/is-goog...
Does anyone care that the head of the art history department did his phd on Picasso, taught two full semester courses on him and I know more about Picasso’s life than probably 99.9% of the general population?
Doubtful…
But I do know things that never made it into the search engines (I’ve looked) and this knowledge is locked up in a (probably) retired professor and the few percent of students who actually listened to his lectures twenty-something years ago.
Now everyone assumes that if it isn’t in google it doesn’t exist and have very little idea how to find information outside of what’s contained in its “record of all things”.
Depends on what the difficulty is. I remember when I started to learn programming back in 1991, I was learning from books. These books were very expensive and I could only afford a few of them. I didn't have access to a library because I wasn't in any school or university. I was a musician with a side job at a theatre box office.
My main problem was that many of the explanations and examples in my programming books left some questions unanswered. It was even worse with maths, which I also tried to learn at the time. I got stuck a lot. If I'd had access to something like stackoverflow or Google, I could have progressed much much faster.
I do think that getting stuck and being forced to figure something out the hard way is sometimes useful. But more often than not I didn't get stuck because the material was actually difficult or because I was solving difficult problems. I got stuck because the one and only explanation available to me was imprecise or incomplete. All I needed was the same thing explained with slightly different words to clear up my misunderstandings.
On the whole I think learning these things today is immeasurably more productive. People are a lot less dependent on luck or money. You don't need to be one of the lucky few who has access to good teachers. You don't have to buy expensive books. Everyone has access to excellent online tutorials, documentation, source code and to thousands of competent professionals willing to help for free. It's absolute heaven. Back then it was hell.
Think about most of the stuff you have really learnt over the course of your life, was it from sitting in a classroom with information being spoon-fed to you, from reading a Wikipedia article?
Or was it because you had to search hard for the information, try and fail before you succeeded?
My bet is that your learning comes from the lather not the former.
Good teachers that I have had would point you in a direction and then give guidance or even question you instead of just telling you the answer.
It’s doing the problem set, but also being able to check the key. Basically a private tutor or TA available at all times.
Without that you can’t get a tight feedback loop.
The most effective use of these systems will not be copying the answer, it’ll be trying to do it and then asking when stuck.
It is absolutely not true for most people. Most people need a human guide to show them the way. I am living this now with my two kids, both of whom need instruction (and also help from mom and dad).
Caveat, of course, this is an extreme power tool vis a vis human development, and as such, can do just as much harm as good depending on it's use. This is why it's so damn important that we keep this stuff open source - if it's behind walled gardens, you end up with _actual, real shadowy cabals effortlessly manipulating all of everybody_, a thing that you may or may not be convinced exists already but guaranteed under those circumstances.
Truth of the matter is that many teachers know the curriculum, but if you ask them a basic question, they won't answer the question with nearly the same depth for example Feynman[1] would have. Now educators with both the ability and depth of knowledge of Feynman is clearly a very limited resource. LLMs may just change that since they can be given the depth with relative ease, the rest of the owl is making them good educators as well.
https://blog.khanacademy.org/harnessing-ai-so-that-all-stude...
It's expensive, it's not prohibitive unless you're not willing to increase education funding at all. Which implies that what we're spending now "should be enough." I don't think we've ever done the basic homework here.
54.5 million school aged children in the US. Average $50,000/yr tutor salary. Total cost per year would be 2.7 trillion.
National Defense is 767 billion. Education is already 677 billion. Net interest is 475 billion. We're already more than half way there, inside of a nation with $25 trillion GDP.
I fail to see anything "prohibitive" about this.
I hate to have to say this but GDP isn’t the absolute upper limit for government spending. I know they tell you that every dollar of government spending is equal to $2 of increased GDP (or some equal bullshit) but that doesn’t mean GDP is an open checkbook for the government to drive the economy. At some point someone has to generate actual value for them to appropriate for their political ends.
This is the only time I’ve ever wanted the AI to generate a reply for me but, alas, that’s not in my current skill set.
Brazil and Zimbabwe found this out the hard way, but a lot of people really don't understand how times can be hard when the numbers keep going up because they neglect to consider inflation and the government has an interest in managing down the real numbers and mumbling something about "greed" while they enact policies that effectively create huge, regressive taxes.
What happens if they have two or more children? What if they want other jobs? What real bearing does this have on the fundamental problem?
> I hate to have to say this but GDP isn’t the absolute upper limit for government spending.
I didn't suggest it was. I'm showing that if you want to have high value education, it would become about 10% of the total economy. Does that seem "prohibitive" to you?
> At some point someone has to generate actual value for them to appropriate for their political ends.
And educating children has no value?
> This is the only time I’ve ever wanted the AI to generate a reply for me but, alas, that’s not in my current skill set.
Well, some educations have no value, I guess.
Obviously, as a cost cutting measure, we will pay $50k per child to each of the parents.
> What real bearing does this have on the fundamental problem?
I don’t actually understand what the problem is we’re trying to solve here, honestly.
Give them all an AI and call it a day.
What is your reasoning for believing this would improve outcomes?
> I don’t actually understand what the problem is we’re trying to solve here, honestly.
Really? We spend a lot on education. Our outcomes are demonstrably sub-optimal. There's no correlation between amount spent in a classroom and classroom outcomes. There's obviously a large flaw in the entire model.
> Give them all an AI and call it a day.
Well.. when we invent an actual AI and not some hallucinatory language model, let me know, until then, you're obviously not serious or willing to play fast and loose with something important.
Hard to believe we're struggling to get this right.