The abacus and calculator analogy is a good one. I've been thinking about ChatGPT in terms of standard language libraries. So far, I think that is what ChatGPT is good for - solving small, self contained problems that are building blocks of larger apps. I'm glad that they are there, and I'm sure ChatGPT will provide a lot of code of this nature.
So I shouldn't be concerned, because I love standard library functions. But for education, I fear that students will be discouraged (or tempted to cheat) when they are learning to code with exercises that can be solved with a simple online query.
I like your extension of the argument towards standard libraries. I guess one way of thinking about this is that the standard library hasn't made programmers obsolete.
Charlie ends his post with an intriguing question - what is the abacus for programming? Some way of reasoning about programming that isn't a black box. Perhaps raising the abstraction level even higher?
Assignments that are like "design an app that makes it really easy to share photos" and then "reason about how you would test that app" and "what countermeasures would you create to prevent abuse" and then have the student(s) come up with specific pieces of implementation of parts of that solution that would fit the allotment of time?
I love those ideas for assignments. Humans will get boxed out of most programming tasks, but the questions that you raise will (probably?) be the last to go
> But for education, I fear that students will be discouraged (or tempted to cheat) when they are learning to code with exercises that can be solved with a simple online query.
Is it cheating when kids learning to read have books read to them?
Is it cheating when teens learning guitar watch YouTube videos of people playing guitar? ...or use Spotify to play music?
Personally, I'm much more excited for how "learning to code" will change given these tools, than I am worried about cheating.
Learning to code is ultimately not about learning a pile of information. It's about learning how to react when faced with a wall. There is definitely something to be said in favour of presenting students with small walls to face before presenting them with large-scale and cognitively overwhelming problems.
The concern here is that most of what we've been using as "small walls to conquer" are trivially solvable without having to go through the problem solving process. But I think this is excessive hand-wringing. StackOverflow is already as disruptive as can be on that regard, but that doesn't prevent a plurality of students from going through the motions for the sake of learning the "craft".
No, those are not cheating. However, it's cheating when your kid types his math problems into Google and just enters them into an autovalidating document. If coding exercises are simple and working code can be acquired brainlessly by typing the problem statement into ChatGPTNext, then yes I'd consider that cheating (both in the direct sense of violating the rules of the course and in the sense of the learner cheating him/herself out of the value of the exercise).
As a side note, the type of math "education" I describe above is worse than useless, as the autovalidator rejects correct answers that aren't entered exactly as the form expects, and has no facility for or requirement to show one's work. It's literally just teaching kids to Google and answer and then shotgun it into the form with random permutations until they get a green checkmark.
AI has really quickly been changing how assignments have to be crafted. In the last couple of years apps like photomath have made the old school approach of teaching math with a lecture on how to do something followed by practice homework problems pretty much useless since students can so easily just have the computer do the problems (complete with showing work for them).
From messing around with ChatGPT a bit, it looks like things like writing essays/book reports/poems for English classes will be going the same way as soon as someone packages it up nicely.
It's going to be a rough transition, but I hope one for the better. I'm not sure about what that will look like for gen-ed classes, but in my CS class, I try to embrace tools like this. Instead of worrying about getting students to memorize boilerplate and algorithms, I try to teach them to research and how to glue things together.
You have a robot that can play the drums okay. Your homework is to record yourself playing a beginner song on the drums. You have the robot do it. You do this for all your homework. You are now in Advanced Drums. Your homework is harder than what the robot can play and you don't know how to play the drums. You have wasted a lot of time.
I spent a good part of last year picking up the abacus and meticulously tracking my progress. I can attest to "continue to visualize the abacus in their mind, improving their skills even when the physical tool is unavailable", but make no mistake, this is a domain-specific skill which has little bearing on mathematical knowledge and intuition per-se (at least not at beginner-intermediate levels).
One useful aspect which is not often discussed, but might actually complement OP's article, is that the mental abacus converts one kind of mental process (numerical perception) to another (visual scratchpad). Kind of like turning text into a QR-code as a more "efficient" 2D-storage, yet the QR-code contains no "text", and you have to convert it back.
This complementary process in the brain becomes interesting when our mental abilities decline or get replaced, and we might as well re-purpose those brain areas using some kind of transfer encoding (e.g. Morse code).
A naive first-guess could be to learn lists of logical challenges and strategies in order to clarify results given by an AI in the fastest way possible. So given a GAN, train the human on A, and leave the G and N to the computer.
It's an interesting observation - that the use of a certain type of tool, while multiplying our productivity, simultaneously leads to atrophy of our ability to do that set of tasks "by hand".
I think the "saving grace" for LLMs in this context is that the user still has to vigorously validate the results.
On a bird's eye perspective... you know as much as I love and enjoy crafting with css/html/javascript ... as I grow older I am not that entertained in having to learn yet another API to do the same thing that we did years ago, only "more elegantly/faster/with less code/etc".
The truth is, if I'm perfectly honest, that to KNOW those APIs in itself, is not real knowledge. It's just empty. Talk to your wife holding your newborn, about the latest tailwind version, or native Javascript additions. Seriously who gives a flying rat's ass?
At this stage of human evolution, we live in a reality bubble where we're told that a life "well lived" is one where you have a "real" job, and one where you "work hard". And your parents tell you to "become somebody"... well maybe it's because they have never know anything else?
Maybe the best thing that will come out of artificial intelligence, is to destroy all this supposed knowledge. All those mountains of nothing.
Imagine you create a new API for a service, and EVEN WItHOUT feeding any examples to the deep learning, it manages to figure out the best way to solve xyz problems using your API, and even highlights inefficiencies in your API. Now we're talking. Why would anybody waste their human potential writing the 1000th tutorial on how to use your API ? That's gonna be replaced by another better API from someone else in 5 days?
I love my work. I love CSS; JS .. I really do but honestly it's also all so empty.
Would I do it all over again? I don't even remember how I wrote in assembler on the Amiga and 486. How I used the copper/blitter etc. All vague distant memories. Everything shifting...
edit: I guess this comes off as a mini rant, simplifying. Not sure what my point is, frustrated with what "learning" means perhaps. Frustrated with the notion that knowledge is repeating stuff.
Actually here's a good question : are teachers really necessary or have we yet to envision a world where children, helped with AI tools, can learn whatever they want - as an innate ability? curiosity?
This resonates with me. Tangentially related — for a few years now I've had my sights set on AI eventually driving the automation of all menial labor needed for human subsistence. Food, water, clothing, housing, healthcare — I see these all as things that every human should have for free. Survival should be free. I've often wondered what amazingly good and horrifyingly bad things humanity will do once nobody has to fight for their basic subsistence any more.
I feel like so much turmoil in the world is caused by people acting irrational because they're afraid of going hungry, becoming homeless, being hurt, etc. A world where that fear is largely gone could be a marvelous utopia where everyone is free to chase after their dreams. It could also result in everyone living in VR 24/7 and our society decaying into hedonistic madness. I think it'll be a little bit of both.
i think that people who believe that such a world is even remotely possible, even with most "menial" tasks automated, are utterly delusional.
resources are finite, even if you're using very efficient fully automated planning & execution systems to utilize said resources. something's got to give, and most likely in such a world it will be the ability to procreate (get neutered, receive shelter, sustenance and healthcare for life).
Considering that most rich first-world countries aren't having enough kids to keep up the rate of replacement, I don't think this is going to be reality. There's a strong correlation between quality of life going up and birthrates going down already, no need to sterilize people.
Also—we're already living in a world where a lot of menial labor is automated. I'm not sure I'd call it "utterly delusional" to see that there's a long history of humans automating tasks and to think that eventually we'll automate everything we need to survive. We've been moving towards that goal for the past 200 years!
automation doesn't solve the resource problem. we can automate seven way to Sunday, it won't change the fact that arable land, fuel, drinking water are all finite and scarce resources, and are set to become even more scarce as the effects of climate change worsen.
I fully subscribe to that goal, but I think we'd need some far broader changes in society to make that a reality. We could already greatly reduce the amount of working hours and distribute essential resources more fairly. But what actually happens is that any increase in efficiently is frantically translated into an increase of output, even if that output is completely unnecessary.
Work as of now is not a necessity to archieve some outcomes, it's what society is organised around in many aspects.
I agree that we could be already making some of these changes, and that the increase in output is definitely gobbling up a bunch of our capacity to do so. That said, we have seen a huge rise in average quality of life over the past 200 years of industrialization. The percentage of people in abject poverty is still much too high, but it's trending downwards [0] and I think AI automation could accelerate that change. I think it may be easier to just scale output so far that it overwhelms human greed — which might sound crazy, but in my opinion is easier than effecting large-scale societal change. We should absolutely be working on societal change as well, though!
We won't stop working unless the AI can give us everything we could desire. With or without UBI we'll be just as busy. Let's say you have lots of time, and super smart AI toys. You got the basics but no luxuries. What would you do? Nothing?
Nope! Which is why I think it would be great—I'd work on things that aren't motivated by needing to pay rent, buy food, provide basic necessities for me and anyone who's dependent on me, etc. I agree that humans are pretty industrious by nature, which is why I think a world where we can be industrious without being forced to be industrious would be pretty cool.
If AI gets to a point where it’s as advanced as you’re describing, 99% of everyone reading this comment will have a job that’s rendered obsolete with 1% of people employed (to manage the AI) leading to a massive concentration of wealth to the very few people necessary to manage the AI and its inputs/outputs.
In such a world, what value do us engineers provide, if any?
Going from their sentiment, probably to be a father. The value you provide as an engineer changes, from being a 40-hr-per-week programmer for a company you may or may not have any personal connection to, to being a father to your child in whatever way you believe the world needs more of, and letting the machine program itself. That we should not get lost devoting a huge chunk of our entire life learning how to maintain an advertising system, just to earn the right to have a place to live in comfort.
Just imagine the imense wealth that will be generated if we can displace 99% of the workforce from what they are doing now. This is something to cheer, not to feel doomed.
That's more a discussion of wealth distribution, systems of government and the flaws of capitalism. Theoretically speaking if you free human labor through the use of machines you're creating wealth because the same or greater output can be achieved with less resources and those resources can be allocated elsewhere. Of course some people believe that machines will replace all jobs eventually but historically that has not been the case. Generally new technology makes some jobs obsolete but humans adapt and find different ways to contribute to society.
We've already replaced 99% of the workforce once. Now instead of subsistence farming, people can live, eat basically any kind of food they want, own a multi-room house, and fly around the world once or twice a year on a vacation, all in exchange for less labor than that same person would have needed to expend to simply not starve a few thousand years ago.
Is that because billionaires "give [you] money out of the goodness of their hearts"? Are you unemployed because all those farming jobs have been eliminated? Obviously not. Rather, the efficiency gains of modern infrastructure have simply made your labor that much more valuable.
Ironically, you're literally describing the 1% lifestyle. (That includes me, but I'm aware that most people don't get to enjoy the great lifestyle you mention.)
?
We're on the verge creating the first trillionaire in the next few decades with our current economy which, gasp, has millions of unemployed people in it.
Most likely they will have to. The governments main role is to make sure the majority of people can live together in a society. If a something as suddenly as 99% of people becoming unemployment overnight the governing bodies will have no choice but to redistribute wealth, kill/enforce any rioters, or dissolve into anarchy.
No, you continue to work for a living, just on something else that's 100x more valuable than before (and you consequently become 100x more wealthy than before).
It's already happened once. 99% of farming jobs that existed a few thousand years ago now long gone. Are we all unemployed and starving because of that? No, in fact we're all far more wealthy than we would have been before all that technological progress was made. We just don't think of it as wealth, and instead complain that the guy next door has more than us and the government ought to do something about that, because he clearly wouldn't be that rich if not for all this darned technological progress taking our jobs.
> No, you continue to work for a living, just on something else that's 100x more valuable than before
This just doesn't make sense. If there's something I can (want to and capable of) work on that's 100x more valuable than what I'm doing now, why isn't somebody hiring me to do that now?
The farming example goes like this -- you may think food is less valuable than your phone because it costs less. But that's just because supply is greater and it costs less to produce, so the cost is lower.
Similarly, if some AI is capable of writing better software than I do now, I don't get to choose another job that pays 100x. It's just that my current job will cost less, and if I continue doing my work, it will be rewarded at 1/100x due to competition from cheap AI. Perhaps I'll be flipping burgers at restaurants and it will be 100x pay compared with writing software. I won't be getting 100x more wealthy though!
> If there's something I can (want to and capable of) work on that's 100x more valuable than what I'm doing now, why isn't somebody hiring me to do that now?
That's like asking why 2000 years ago the farmers didn't just buy a tractor and a combine and do their farming that way, since that would have generated so much more value. The technology doesn't exist right now, we're talking about a scenario where it will.
> Similarly, if some AI is capable of writing better software than I do now, I don't get to choose another job that pays 100x. It's just that my current job will cost less
Correct, assuming that it's only writing software that gets automated, and nothing else. Society as a whole will be wealthier by virtue of being able to afford more software, but not 100x wealthier. But if 99% of everything gets automated, the jobs that remain (and the new ones that get created as a side-effect) will indeed pay 100x more in terms of real purchasing power. Whether you think of that as "this job is 100x more valuable" or "everything else costs 99% less" is just a matter of perspective.
It's clear that the current economic system cannot accommodate a world like that and we either change or live in a world with a handful of trillionaires and billions of desperately poor people.
Lots of people with no jobs, nothing to do, just sitting and taking the UBI? I think people will demand the right to work to support their needs. Even if that means just helping people get land, materials, tools and energy. People can't be passive wards of the state.
Your job won’t be obsoleted by AI even if the AI is literally better at the job than you are. One, that’s not how comparative advantage works, and two, the AI probably costs more than you.
> have we yet to envision a world where children, helped with AI tools, can learn whatever they want - as an innate ability? curiosity?
If there's any danger of a world like this happening, rent-seekers will step in to prevent it. We have the internet, and somehow educational materials for every single trade and science aren't freely available and neatly organized on it. In fact, we spend almost all of our worry on the internet trying to prevent information from flowing through it.
We invented a world brain, and spend all of our time installing and shoring up virtual amyloid plaques on it.
Sorry, that sounds awfully nihilist to me. No, I'm not terribly excited in becoming an expert for the latest API or framework either, but I'm very interested in understanding how the world around us works. There are also some skills I'm proud to have and others I'd like to have.
I share the worry of the article that AI might become the ultimate competitive artifact: Make us able to realise any idea on a whim, while at the same time making it actually harder to gain a deeper understanding of something or acquire a skill. In that world, everyone will basically become a pure consumer.
It's not the only path a world with AI could take, but it's one that I worry is quite likely. And I don't think this would make anything better.
At least I don't support the notion that as soon as you have a child, the entire rest of the world stops being relevant. First of all, that sounds horrible, secondly, not everyone will have kids.
> a world where children, helped with AI tools, can learn whatever they want - as an innate ability? curiosity?
That would be a better vision. I read in some other thread that at some point we all might get personal assistants that go with us through life like some non-imaginary imaginary friend. I think this could actually be awesome.
We spend a lot of our time in life being rewarded for showing up on time, following instructions, and filling our time keeping busy. Learning yet another way to accomplish the same task is right in the middle of this.
Then there’s the real results-oriented perspective. Where we choose a goal, a good goal to make our world better for ourselves and others. Where you measure progress towards that goal and not just time and money spent.
And so what if all the drudgery disappeared tomorrow? If we could leave a thousand overcomplicated methods behind and just speak what we want and have it created.
Sounds wonderful to me and I would leave my tools in the dusts of history.
some things in life are deeper, and I think the electronics/computing era was half pragmatism half 'what-if' and we reached the what-if, it's everywhere, it's near omnipotent.. the itched is scratched, and while it's cool, the impact is not all glorious, other factors (sometimes archaic ones) weigh in our societies (the current state says it all) and our lives.
but back to computing, that's why I'm aiming a lot more on some mathy topics, it has a longer life time, more poetry. That and cosmology like fields. There's something strangely peaceful about wondering about the universe.
ps: other fields of value: biology/medicine, pragmatic sociology (how to help groups living better), and agro/ecology.
I derive zero intrinsic enjoyment out of learning frameworks or APIs, the only reason that I do so is that they help me achieve a vision of something that I want to build. Yesterday I cobbled together a Kotlin app with the help of chat GPT that when a phone is detected in freefall using the accelerometer it would start screaming. I've never used kotlin before, and I was able to put it together in a single day.
If chat GPT or some AI better facilitates the ability for me to build and create new things that I'm just as happy.
The conclusion that high level programming languages are "complementary" while LLMs are "competitive" seems very shaky to me. If we had never invented high level languages, and instead used LLMs to assist writing and debugging assembly code, I think we'd all be a lot better at reading assembly.
I think the real effect on programming is to reduce the importance of mechanical and rote skills such as typing fast and having a photographic memory for standard library functions and algorithm patterns.
If your main skill is that when given a detailed description of something to implement, you can bang out working code at 100 WPM, then LLMs are kind of making that obsolete.
If your main skill is balancing requirements, making judgement calls, and reviewing code for correctness, then LLMs are going to help you work a lot faster.
i think in longer terms. step one is what you mentioned. that's what people thought when computers came along and they said it will probably only be used to assist some scientists with complex calculus.
step two is to automate the business analyst then the architect.
step three might be to directly render whatever requirements the customer or enduser gives the LLM. and i bet this comes in under 10 years. Basically what StableDiffusion does right now. No Graphic designer needed.
And imagine what will happen, if you allow an LLM 10 years in the future to improve it's own architecture and training data.
And now imagine if you allow that thing to control chip factories and assembly lines to improve it's hardware...
The hype over this is bordering the ridiculous claims made by initial AI researchers in the 50s. So chatGTP is going to be running its own factories in a decade. I guess we should make sure it doesn’t overvalue paper clip production. Or cyborgs and time travel.
There’s a big difference between using a tool (all AI to this point), and the tool becoming its own agent in the world, pursuing goals independent of humans, which is what you get in scifi.
ChatGPT responds at much faster than 100WPM, but typing fast is still going to be useful/important, because it's still important to express what you want with clarity to an LLM
GPT3 could be a legit game changer for teaching CS. The first time I tried self-learning programming I gave up because I often got stuck on random things that a tutor/mentor could have helped me solve in 5 seconds instead of hours of Googling.
Some sort of fine-tuned model that can analyze why your code isn't working for practice problems and give hints and suggestions rather than just giving you the solution. Each student could effectively have their own personal tutor
obviously people are going to use "unlocked" GTP3 to do their homework, but kids do that today as well. People who want to actually learn will be helped massively by tools built on LLMs
And there’s so much attention on “oh no, take-home assignments are impossible now” but if so, I see the education world adapting. Instead of writing an essay about Napoleon, a bot will interview you like an oral exam — a far superior evaluation, but when done by humans there was just never enough to time in their day to do them regularly and for everyone. Instead of a homework assignment done behind closed doors and then tossed over the fence for the TA to grade, the bot pair programs with you and helps you through the places you’re getting stuck.
A wonderful outcome if that level of adaption occurs. I am more worried that the Luddite mentality will get these models banned in (primary school) educational settings.
Yeah, I could see a position of, “How will people express themselves if they don’t practice writing essays?” but I could see in five years the primary two ways of doing that becoming (1) Machine-assisted essay writing, and (2) People with ideas explaining them to an AI model through conversation, and then the model going out and explaining it en masse to others. Maybe even some of those others will challenge points or extend the ideas, and the tech can distill all this feedback and critique into a summary that they take back to the author.
The problem with using GPT for education is it isn't an intelligence, it is just a fancy pattern matcher. It is as likely to give you a wrong or bad answer as a right answer. A student wouldn't be able to tell the difference, so they might learn the wrong stuff. It is handy for people with the experience to tell the difference though.
Only in closed-book mode. If you open the books and let it reference the materials it doesn't make stupid mistakes like that. What the model needs is a search engine and a curated index of materials.
But how much different is that from googling and getting a crappy stack-overthrow answer, or a poorly written blogpost? In the areas I am an expert in, the information I find is often incorrect, insufficient, inaccurate, invalid and often downright incredibly shitty. Still better than nothing?
A bigger problem is that GPT depends on us all writing and publishing answers and examples, which we often do as part of our discovery or learning process. If we are just given answers, perhaps we won't publish answers, and then GPT will not have the material it requires for it to give its answers. Also we are mostly motivated to write because we have an audience: however with the intercession of GPT, the audience disappears and there is less motivation to publish.
Is chatgpt a weak learning tool? ,that is, does it have the capacity to begin with basic fa ts that provides a probability above Zero about knowing something and then combine the database encoded knowledge to guarantee a lower bound of general knowledge?
I’ve also been using ChatGPT like crazy since it came out. I am a tech CEO and occasionally write code to automate various CEO things or to prototype technical ideas. Although I trained as a software engineer years ago, I can’t afford to spend a lot of time learning new frameworks or languages just to get a simple prototype written up.
ChatGPT is very helpful in getting past the learning curve. For example, I wanted to write a script that pulls data from RabbitMQ, does some text manipulation, and sends the result to DynamoDB. And, for fun, I wanted to write it in Rust. ChatGPT quickly gave me a working template and I was on my way.
I can’t wait for ChatGPT to reach 1.0 as a paid service. I will gladly pay to accelerate the boilerplate phase of my projects.
Copilot is AI IntelliSense. You still have to write code.
ChatGPT is an AI programmer (of limited skill, but still). You describe what you want in plain English and get a runnable program out the other side. You may or may not have to modify this program. It can also make changes to the program based on plain English descriptions.
Nobody uses an abacus anymore. I use a calculator (well, Mathematica or scipy) to effortlessly solve problems that would be impossible to solve by hand. I appreciate, but don’t particularly understand or care about, the many decades of numerical analysis research that went into making scipy algorithms robust and efficient.
ChatGPT is the most exciting thing to come out of CS in a long while. Forget all of the debate and angst about whether ChatGPT is “really” an AI and what its “real” capabilities are (and I squarely blame the AI community’s recent history of brazen overhype and self-promotion for this)—-it’s a red herring. Call it the most powerful Internet search engine ever invented, if you want. That alone will change the world.
It will trivialize a lot of tasks that are currently laborious and burdensome. The hand-wringing I’m seeing about this reminds me a lot of the rise of the Internet in the 90s, or Wikipedia in the 00s… yes, we will need to rethink how we write software and we will need to rethink how we teach students. But getting hung up on the dangers of ChatGPT as a “competitive” too is to cling a bit too tightly to old paradigms about of how humans leverage technology that have never been rigid, even in our lifetime.
Will chatGPT be free to use as a search like Google is, and why hasn’t Google leveraged its own language model? What happens to websites when nobody visits anymore because their info is easily available in the search response?
That will be interesting to see. I bet by 2030 few people will be seeing ads or websites unless they want to, most will get polite and helpful answers based on language models that use search engines under the hood.
We'll eventually have our own models running on our computers, in privacy, and maybe a small index of text (few TB) for reference checking. This will create a new private space of creativity where we can't be tracked and advertised to.
ChatGpt doesn't augment education, if replaces it. A version that is 99% correct on a subject will be enough to replace the teacher. It is endlessly patient, unlike any teacher. Coupled with high quality video courses, i can see children above a certain age using it to learn and explore science,math, languages and literature
That would still be "education", so I’m not sure what part you mean it will replace. Do you mean it will replace educators? If so, who exactly would be creating these “high level” video courses?
Also who would provide the curriculum? Someone would have to tell the child: “This week, have a conversation with ProfessorGPT on the following topic. At the end of a week, submit an essay (generated by ProfessorGPT) that shows a summary of your discussion, and include a transcript of the entire session.”
You’re also assuming a socratic approach to education works for every child. I’m not sure that’s universally true.
I was nodding along in the introduction, rolling my eyes in cognitive artifacts, and sighing in the conclusion.
It seems like math educators still loathe students use calculators, and I find the reasoning has major blindspots. Competitive artifacts abound in society, and we are better for it. If I am in a situation where I do not have immediate access to a calculator, I am not in a situation where I need to do quick math. In 20 years, if we're still asking what year WW1 started on history tests, then we utterly failed.
We absolutely need to find a way to teach programming in a world of AI. We also need to find a way to teach art in a world of AI. We'll need to find a way to teach everything in a world of AI. What we don't need to hold onto is this mistaken belief that we need to somehow teach people how to get by without it. It's like telling someone they need to learn all of a python framework before they're allowed to use it...utterly defeats the point of its existence.
77 comments
[ 5.4 ms ] story [ 60.7 ms ] threadSo I shouldn't be concerned, because I love standard library functions. But for education, I fear that students will be discouraged (or tempted to cheat) when they are learning to code with exercises that can be solved with a simple online query.
Charlie ends his post with an intriguing question - what is the abacus for programming? Some way of reasoning about programming that isn't a black box. Perhaps raising the abstraction level even higher?
Assignments that are like "design an app that makes it really easy to share photos" and then "reason about how you would test that app" and "what countermeasures would you create to prevent abuse" and then have the student(s) come up with specific pieces of implementation of parts of that solution that would fit the allotment of time?
Is it cheating when kids learning to read have books read to them?
Is it cheating when teens learning guitar watch YouTube videos of people playing guitar? ...or use Spotify to play music?
Personally, I'm much more excited for how "learning to code" will change given these tools, than I am worried about cheating.
The concern here is that most of what we've been using as "small walls to conquer" are trivially solvable without having to go through the problem solving process. But I think this is excessive hand-wringing. StackOverflow is already as disruptive as can be on that regard, but that doesn't prevent a plurality of students from going through the motions for the sake of learning the "craft".
As a side note, the type of math "education" I describe above is worse than useless, as the autovalidator rejects correct answers that aren't entered exactly as the form expects, and has no facility for or requirement to show one's work. It's literally just teaching kids to Google and answer and then shotgun it into the form with random permutations until they get a green checkmark.
AI has really quickly been changing how assignments have to be crafted. In the last couple of years apps like photomath have made the old school approach of teaching math with a lecture on how to do something followed by practice homework problems pretty much useless since students can so easily just have the computer do the problems (complete with showing work for them).
From messing around with ChatGPT a bit, it looks like things like writing essays/book reports/poems for English classes will be going the same way as soon as someone packages it up nicely.
It's going to be a rough transition, but I hope one for the better. I'm not sure about what that will look like for gen-ed classes, but in my CS class, I try to embrace tools like this. Instead of worrying about getting students to memorize boilerplate and algorithms, I try to teach them to research and how to glue things together.
Relevant XKCD: https://xkcd.com/1988/
One useful aspect which is not often discussed, but might actually complement OP's article, is that the mental abacus converts one kind of mental process (numerical perception) to another (visual scratchpad). Kind of like turning text into a QR-code as a more "efficient" 2D-storage, yet the QR-code contains no "text", and you have to convert it back.
This complementary process in the brain becomes interesting when our mental abilities decline or get replaced, and we might as well re-purpose those brain areas using some kind of transfer encoding (e.g. Morse code).
A naive first-guess could be to learn lists of logical challenges and strategies in order to clarify results given by an AI in the fastest way possible. So given a GAN, train the human on A, and leave the G and N to the computer.
I think the "saving grace" for LLMs in this context is that the user still has to vigorously validate the results.
The truth is, if I'm perfectly honest, that to KNOW those APIs in itself, is not real knowledge. It's just empty. Talk to your wife holding your newborn, about the latest tailwind version, or native Javascript additions. Seriously who gives a flying rat's ass?
At this stage of human evolution, we live in a reality bubble where we're told that a life "well lived" is one where you have a "real" job, and one where you "work hard". And your parents tell you to "become somebody"... well maybe it's because they have never know anything else?
Maybe the best thing that will come out of artificial intelligence, is to destroy all this supposed knowledge. All those mountains of nothing.
Imagine you create a new API for a service, and EVEN WItHOUT feeding any examples to the deep learning, it manages to figure out the best way to solve xyz problems using your API, and even highlights inefficiencies in your API. Now we're talking. Why would anybody waste their human potential writing the 1000th tutorial on how to use your API ? That's gonna be replaced by another better API from someone else in 5 days?
I love my work. I love CSS; JS .. I really do but honestly it's also all so empty.
Would I do it all over again? I don't even remember how I wrote in assembler on the Amiga and 486. How I used the copper/blitter etc. All vague distant memories. Everything shifting...
edit: I guess this comes off as a mini rant, simplifying. Not sure what my point is, frustrated with what "learning" means perhaps. Frustrated with the notion that knowledge is repeating stuff.
Actually here's a good question : are teachers really necessary or have we yet to envision a world where children, helped with AI tools, can learn whatever they want - as an innate ability? curiosity?
I feel like so much turmoil in the world is caused by people acting irrational because they're afraid of going hungry, becoming homeless, being hurt, etc. A world where that fear is largely gone could be a marvelous utopia where everyone is free to chase after their dreams. It could also result in everyone living in VR 24/7 and our society decaying into hedonistic madness. I think it'll be a little bit of both.
resources are finite, even if you're using very efficient fully automated planning & execution systems to utilize said resources. something's got to give, and most likely in such a world it will be the ability to procreate (get neutered, receive shelter, sustenance and healthcare for life).
Also—we're already living in a world where a lot of menial labor is automated. I'm not sure I'd call it "utterly delusional" to see that there's a long history of humans automating tasks and to think that eventually we'll automate everything we need to survive. We've been moving towards that goal for the past 200 years!
Work as of now is not a necessity to archieve some outcomes, it's what society is organised around in many aspects.
[0]https://en.wikipedia.org/wiki/Extreme_poverty#/media/File:Wo...
If AI gets to a point where it’s as advanced as you’re describing, 99% of everyone reading this comment will have a job that’s rendered obsolete with 1% of people employed (to manage the AI) leading to a massive concentration of wealth to the very few people necessary to manage the AI and its inputs/outputs.
In such a world, what value do us engineers provide, if any?
Is that because billionaires "give [you] money out of the goodness of their hearts"? Are you unemployed because all those farming jobs have been eliminated? Obviously not. Rather, the efficiency gains of modern infrastructure have simply made your labor that much more valuable.
That’s literally the plot of Atlas Shrugged, not real life.
It's already happened once. 99% of farming jobs that existed a few thousand years ago now long gone. Are we all unemployed and starving because of that? No, in fact we're all far more wealthy than we would have been before all that technological progress was made. We just don't think of it as wealth, and instead complain that the guy next door has more than us and the government ought to do something about that, because he clearly wouldn't be that rich if not for all this darned technological progress taking our jobs.
This just doesn't make sense. If there's something I can (want to and capable of) work on that's 100x more valuable than what I'm doing now, why isn't somebody hiring me to do that now?
The farming example goes like this -- you may think food is less valuable than your phone because it costs less. But that's just because supply is greater and it costs less to produce, so the cost is lower.
Similarly, if some AI is capable of writing better software than I do now, I don't get to choose another job that pays 100x. It's just that my current job will cost less, and if I continue doing my work, it will be rewarded at 1/100x due to competition from cheap AI. Perhaps I'll be flipping burgers at restaurants and it will be 100x pay compared with writing software. I won't be getting 100x more wealthy though!
That's like asking why 2000 years ago the farmers didn't just buy a tractor and a combine and do their farming that way, since that would have generated so much more value. The technology doesn't exist right now, we're talking about a scenario where it will.
> Similarly, if some AI is capable of writing better software than I do now, I don't get to choose another job that pays 100x. It's just that my current job will cost less
Correct, assuming that it's only writing software that gets automated, and nothing else. Society as a whole will be wealthier by virtue of being able to afford more software, but not 100x wealthier. But if 99% of everything gets automated, the jobs that remain (and the new ones that get created as a side-effect) will indeed pay 100x more in terms of real purchasing power. Whether you think of that as "this job is 100x more valuable" or "everything else costs 99% less" is just a matter of perspective.
We would do systems engineering at the domain level. And when AI can do that, we don't have to worry about jobs anymore.
Who is telling you that??
If there's any danger of a world like this happening, rent-seekers will step in to prevent it. We have the internet, and somehow educational materials for every single trade and science aren't freely available and neatly organized on it. In fact, we spend almost all of our worry on the internet trying to prevent information from flowing through it.
We invented a world brain, and spend all of our time installing and shoring up virtual amyloid plaques on it.
That's a cute naive sentiment, that kids want to learn on their own. You have to make them want to learn, that's the role of a teacher.
I share the worry of the article that AI might become the ultimate competitive artifact: Make us able to realise any idea on a whim, while at the same time making it actually harder to gain a deeper understanding of something or acquire a skill. In that world, everyone will basically become a pure consumer.
It's not the only path a world with AI could take, but it's one that I worry is quite likely. And I don't think this would make anything better.
At least I don't support the notion that as soon as you have a child, the entire rest of the world stops being relevant. First of all, that sounds horrible, secondly, not everyone will have kids.
> a world where children, helped with AI tools, can learn whatever they want - as an innate ability? curiosity?
That would be a better vision. I read in some other thread that at some point we all might get personal assistants that go with us through life like some non-imaginary imaginary friend. I think this could actually be awesome.
We spend a lot of our time in life being rewarded for showing up on time, following instructions, and filling our time keeping busy. Learning yet another way to accomplish the same task is right in the middle of this.
Then there’s the real results-oriented perspective. Where we choose a goal, a good goal to make our world better for ourselves and others. Where you measure progress towards that goal and not just time and money spent.
And so what if all the drudgery disappeared tomorrow? If we could leave a thousand overcomplicated methods behind and just speak what we want and have it created.
Sounds wonderful to me and I would leave my tools in the dusts of history.
some things in life are deeper, and I think the electronics/computing era was half pragmatism half 'what-if' and we reached the what-if, it's everywhere, it's near omnipotent.. the itched is scratched, and while it's cool, the impact is not all glorious, other factors (sometimes archaic ones) weigh in our societies (the current state says it all) and our lives.
but back to computing, that's why I'm aiming a lot more on some mathy topics, it has a longer life time, more poetry. That and cosmology like fields. There's something strangely peaceful about wondering about the universe.
ps: other fields of value: biology/medicine, pragmatic sociology (how to help groups living better), and agro/ecology.
I derive zero intrinsic enjoyment out of learning frameworks or APIs, the only reason that I do so is that they help me achieve a vision of something that I want to build. Yesterday I cobbled together a Kotlin app with the help of chat GPT that when a phone is detected in freefall using the accelerometer it would start screaming. I've never used kotlin before, and I was able to put it together in a single day.
If chat GPT or some AI better facilitates the ability for me to build and create new things that I'm just as happy.
Maybe some sort of radar ping... vigilance mode enabled...
I think the real effect on programming is to reduce the importance of mechanical and rote skills such as typing fast and having a photographic memory for standard library functions and algorithm patterns.
If your main skill is that when given a detailed description of something to implement, you can bang out working code at 100 WPM, then LLMs are kind of making that obsolete.
If your main skill is balancing requirements, making judgement calls, and reviewing code for correctness, then LLMs are going to help you work a lot faster.
step two is to automate the business analyst then the architect.
step three might be to directly render whatever requirements the customer or enduser gives the LLM. and i bet this comes in under 10 years. Basically what StableDiffusion does right now. No Graphic designer needed.
And imagine what will happen, if you allow an LLM 10 years in the future to improve it's own architecture and training data.
And now imagine if you allow that thing to control chip factories and assembly lines to improve it's hardware...
Some sort of fine-tuned model that can analyze why your code isn't working for practice problems and give hints and suggestions rather than just giving you the solution. Each student could effectively have their own personal tutor
obviously people are going to use "unlocked" GTP3 to do their homework, but kids do that today as well. People who want to actually learn will be helped massively by tools built on LLMs
A bigger problem is that GPT depends on us all writing and publishing answers and examples, which we often do as part of our discovery or learning process. If we are just given answers, perhaps we won't publish answers, and then GPT will not have the material it requires for it to give its answers. Also we are mostly motivated to write because we have an audience: however with the intercession of GPT, the audience disappears and there is less motivation to publish.
ChatGPT is very helpful in getting past the learning curve. For example, I wanted to write a script that pulls data from RabbitMQ, does some text manipulation, and sends the result to DynamoDB. And, for fun, I wanted to write it in Rust. ChatGPT quickly gave me a working template and I was on my way.
I can’t wait for ChatGPT to reach 1.0 as a paid service. I will gladly pay to accelerate the boilerplate phase of my projects.
ChatGPT is an AI programmer (of limited skill, but still). You describe what you want in plain English and get a runnable program out the other side. You may or may not have to modify this program. It can also make changes to the program based on plain English descriptions.
ChatGPT is the most exciting thing to come out of CS in a long while. Forget all of the debate and angst about whether ChatGPT is “really” an AI and what its “real” capabilities are (and I squarely blame the AI community’s recent history of brazen overhype and self-promotion for this)—-it’s a red herring. Call it the most powerful Internet search engine ever invented, if you want. That alone will change the world.
It will trivialize a lot of tasks that are currently laborious and burdensome. The hand-wringing I’m seeing about this reminds me a lot of the rise of the Internet in the 90s, or Wikipedia in the 00s… yes, we will need to rethink how we write software and we will need to rethink how we teach students. But getting hung up on the dangers of ChatGPT as a “competitive” too is to cling a bit too tightly to old paradigms about of how humans leverage technology that have never been rigid, even in our lifetime.
We'll eventually have our own models running on our computers, in privacy, and maybe a small index of text (few TB) for reference checking. This will create a new private space of creativity where we can't be tracked and advertised to.
Also who would provide the curriculum? Someone would have to tell the child: “This week, have a conversation with ProfessorGPT on the following topic. At the end of a week, submit an essay (generated by ProfessorGPT) that shows a summary of your discussion, and include a transcript of the entire session.”
You’re also assuming a socratic approach to education works for every child. I’m not sure that’s universally true.
It seems like math educators still loathe students use calculators, and I find the reasoning has major blindspots. Competitive artifacts abound in society, and we are better for it. If I am in a situation where I do not have immediate access to a calculator, I am not in a situation where I need to do quick math. In 20 years, if we're still asking what year WW1 started on history tests, then we utterly failed.
We absolutely need to find a way to teach programming in a world of AI. We also need to find a way to teach art in a world of AI. We'll need to find a way to teach everything in a world of AI. What we don't need to hold onto is this mistaken belief that we need to somehow teach people how to get by without it. It's like telling someone they need to learn all of a python framework before they're allowed to use it...utterly defeats the point of its existence.