In most cases the courses are fine, but the examination probably isn't testing anything useful if it can be cheated with tools like ChatGPT. For example any maths test which only tests students on problems which are easily solvable by a calculator isn't testing any skill of real world use. But this doesn't mean maths itself isn't valuable, just that the test isn't.
In some cases like programmers and writers, ChatGPT might actually be eroding the value the current curriculum. For example, I'm not sure if there's much value in learning things like basic SQL queries anymore which was something I had to learn when studying computer science. ChatGPT is great at writing SQL queries if you're able to tell it exactly what you need.
I really disagree, a major function of education is to teach concepts before introducing tools. I took a physics class in undergrad that required a graphing calculator, does that mean I wasted years learning all of the mathematical concepts that could be performed by a graphing calculator?
Having an understanding of what's going on is so important, especially when something is broken. If we encourage students to start with AI and not learn the fundamentals themselves, they won't wind up learning how to identify root cause when the tool causes an issue (and ChatGPT is by no means perfect).
If they successfully use ChatGTP to do this, they likely have engaged with the text.
Using ChatGPT to generate something that you can't independently verify is a terrible idea. Most of the time it'll be fine, but when it's not you need to be able to recognize the flaws. Hopefully you at least test it before pushing, but if you don't know how to write the query you also don't know its failure modes, so your tests will likely miss important edge cases.
AI tools are great for speeding up processes you can do by hand, but because they're all statistics you can't count on them to do it alone: some percentage of the time they will inevitably be wrong.
EDIT: Note that this stands in contrast to a wholly deterministic tool like a calculator, which in most circumstances doesn't need to be second guessed.
It depends. A calculator can calculate things I'd struggle to independently verify with a pen and paper.
I think I wasn't clear enough and you might be taking what I said to it's extreme. I'm not suggesting there is zero value in knowing SQL, I'm saying the value of learning it is now much lower than when I learnt it. While it might be helpful to have a good understanding of SQL to verify a query yourself (especially where performance is a concern), in most cases this likely isn't necessary. For your average select query with a couple of joins and a where clause ChatGPT can generate that and you can verify it at least works by running it.
But maybe SQL isn't the best example. It just came to my mind because I was writing a spatial query for MariaDb recently and I couldn't get the results I wanted after a few Google searches, but ChatGPT got it right first time. It seems to me that query languages have a lot of properties which make ChatGPT very good at writing them, but there are probably other skills which ChatGPT more clearly erodes the value of.
I would imagine writing is similar - the endless essays full of bullshit I pumped out in high school were obviously pointless. But being forced to write much earlier in school was super foundational for having the skill in the first place.
As to the SQL example, I also think it’s flipped - learning the basic sql queries is probably the most valuable as it’s a foundation.
True in university, but memorizing multiplication tables and learning the mechanics of long division and multiplication are important parts or math education for kids. Most of school is not strictly for "real world" skills anyway (even if doing basic math unaided is almost as close as we come in school to teaching a real world skill).
Same goes for say, basic reading comprehension and summarization.
I don't buy your argument
At the same time it is a completely different mode to operate. Writing and editing SQL while you are trying to solve a problem will lead you to a different way of thinking, iterating, maybe even induce fewer context switches.
Using ChatGPT can be similar to delegating those type of tasks to a colleague. There are upsides and downsides to this but never underestimate the benefits of strengthening your own abilities.
--- AMC 8 The only materials that students are allowed to have on themselves during the competition are writing utensils, blank scratch paper, rulers, and erasers. NO CALCULATORS OR PHONES AND SIMILAR ELECTRONIC DEVICES OF ANY KIND ARE ALLOWED. No questions require the use of a calculator.
AMC 10/12 - AIME The only materials that students are allowed to have on themselves during the competition are writing utensils, blank scratch paper, rulers, compasses, and erasers. NO CALCULATORS OR PHONES AND SIMILAR ELECTRONIC DEVICES OF ANY KIND ARE ALLOWED. No questions require the use of a calculator. --
Mathcounts[2] is kinder, but only a little bit - "Calculators are not permitted in the Sprint and Countdown Rounds, but they are permitted in the Target, Team and Tiebreaker Rounds". There are 500+ chapter rounds every year, plus 56 state rounds, & each school sends upto 12 contestants, so again a very large number of students compete in these exams with no calculator.
Note when they say "No questions require the use of a calculator" - that's quite a dubious claim. For eg. the question which is bigger 9^10 or 10^9 ? With a calculator it becomes trivial. Without, you take logs to base 10 and do some simple arithmetic by hand. For most trig problem sets, you'd have to know the standard sines & cosines. With a calculator that becomes moot.
[1] https://www.maa.org/math-competitions/amc-policies [2] https://www.mathcounts.org/programs/official-rules-procedure...
Prompt engineering is a skill I'm seeing a need for
It tends to land in the right ballpark and capture the tone of a thing.
It's too bad that it produces such plausible-sounding answers while being subtly wrong. But it can certain help people to expand their knowledge. I don't think it's a replacement for a comprehensive education on a topic, but can help people get closer.
It cooler if it were an “auto-answer” tool for Stack Overflow. That way I could still get instant answers, and the community could asynchronously validate the answers.
"Schools are banning chatGPT but children won't let schools tamper with their education"
> While the tool may be able to provide quick and easy answers to questions, it does not build critical-thinking and problem-solving skills, which are essential for academic and lifelong success
An early example: http://files.righto.com/calculator/sinclair_scientific_simul... (see Bugs and limitations).
More contemporary (I think): https://apcentral.collegeboard.org/courses/resources/example...
Granted though, ChatGPT is confidently and convincingly wrong more often. I expect that will improve, but in the meantime I think it's an opportunity to question and evaluate the results rather than ban outright. That's a sticking plaster that hopefully needn't stay on for long.
I'd say ChatGPT is a bit better as a tool than copying from a colleague, just because you still have to double check for correctness. However, does it rob students of that opportunity to think deeply and therefore learn? I'm not sure to what extent, but I'd imagine that there's an impact to _some_ extent.
ChatGPT is also going to get better, and will eventually be pretty accurate (or, at least as accurate as a student of a subject may be). My stance is that in school, you should be learning to think instead of memorising. If the goal is to memorise, however, then use ChatGPT isn't that harmful (you'll still fail for memorising the wrong information).
Overall I'd treat ChatGPT as a tool, and give students the facts: first gain a deep understanding of a subject so you can learn to verify, then use the tool later in life.
I'm also mindful of the fact that if an AI is not guaranteed to be correct, which is the case right now, then students now have an incentive to _understand_ the subject so they can verify, which might actually be a net good?
- Google: How do I ... - visit stack overflow - copy/paste - code generated.
There's more steps, but how is this different? Chat GPT though will go the extra mile and actually EXPLAIN what each bit does usually. It's not always accurate but neither is stackoverflow.
That’s what « generating » means, in that context.
One weird trick is to use this text to search the web. Because it is already in the ballpark it will be a very good query. The model would bullshit "The height of Everest is 8,230 m", you take this string and search it, find "height of Everest is 8,849 m" -> correct the original response.
Now the problem is, how do you decide what information to trust in search?
AI needs a big push to index and do consistency checks for all facts. Let's have the model write a billion wiki entries and knowledge base concepts. Check for support, consistency, competing explanations - everything should be in there, we don't decide what is true. Then a language model can say when a topic is controversial, or when it doesn't know something.
Of course the goal is to find the truth, but we know it is going to be tricky. In the meantime we can have models that know when they don't know, or when the information is not certain.
This system can’t prove its own work (at least not yet) before publishing the results. It just publishes, which is similar to Wikipedia and Google.
Assuming the teacher is doing their job, it should be really easy to make sure nobody is cheating using machine learning.