Hey, there! Log in / Register
BU computer-science department devises policy for the use of AI in classwork
By adamg on Mon, 04/10/2023 - 9:20am
The Daily Free Press reports the Boston University Center for Computing and Data Science's new policy is that students can use AI "if they properly credit the tool and include an appendix showing where and how AI tools were used." It does quote one lecturer, though, who feels AI is sort of like calculators were back in the day: It's OK for advanced students who already have the basics down, but intro students shouldn't use it to get answers instead of learning those basics first.
Neighborhoods:
Topics:
Free tagging:
Ad:
Support Universal Hub
Help keep Universal Hub going. If you like what we're up to and want to help out, please consider a (completely non-deductible) contribution.
Comments
Seems reasonable. Honestly,
Seems reasonable. Honestly, using ChatGPT for help on a coding project really isn't all that different from just looking things up on Stack Overflow.
Stack Overflow
I’ve never used ChatGP. When you ask it, “how do I do this?”, does it answer “don’t do that”, or “only a moron would want to do that”? There’s a lot of that on Stack Overflow.
Looks to me like it's getting better, fast. The
recent GPT-4 release got 1410 on its SATs and 4s and 5s on its AP tests. It may not be much at coding at the moment, but I wouldn't look away for long.
It's better than Stack Overflow
I was working with it this weekend to improve some SQL queries, and I could feed it my code and explain what I wanted (e.g., could I make this more efficient?). When I asked it to explain why certain things would be better, it did.
(I also tested it against data to make sure it was right, but I'm such a noob SQL person that my cases are not exactly unique problems to solve.)
ChatGPT is like an assistant. Not an expert, but good enough to give you ideas you can improve. If you're one of those people who figures things out by talking out loud, debating with someone, or explaining something, it's great.
But it's all in what you feed it
ChatGPT is all what you feed it for knowledge. You feed it full of wrong crap, its going to spew that back at you. Yes it learns how to parse all that and fix that.
I'd just be wary of people relying on this as a defacto way to information.
Haven't we learned this from the internet already? Anyone can load anything on here... ChatGPT I have friends making porno fetish chat bots and I hear rumblings about maga loading this up with crap too.
So its all in what you feed it.
Full Disclosure: I don't like ChatGPT, I think it will cause great damage to our society that will outweigh the good points of it. . o O (just like social media) O o .
It gives a lot of confidently wrong responses
So, a great replacement for Stack Overflow. ;-)
No, seriously, I gave it a tiny piece of code and asked it to find the bug, and not only did it fail to find it, it actually asserted that the bug didn't exist and then made up three more, and finally gave me some more-broken code.
I mean, no surprise: A jumped-up autocomplete isn't great at programming. But people will happily take the crap it spits out.
This is OK for regular bugs, but when it comes to security (non-obvious, but exploitable bugs), it gets a lot more exciting.
EDIT: Here are some examples: https://www.brainonfire.net/blog/2023/02/21/chatgpt-hallucinating-bugs/
Further to that point
If you're taking ChatGPT code to do anything even semi-secure with it, you're a fool.
Where's the ethical agreement from OpenAI that they won't *intentionally* insert exploitable security loopholes in ChatGPT answers to such coding questions so that OpenAI know they can exploit whatever you deploy using code from ChatGPT?
Do you find that agreement anywhere? If so, do you find it to be legally binding?
Are you happy to spend whatever money you have left chasing them for violating that agreement in order to recover some money after everyone stops buying your product and you have no more customers because it was easily exploitable?
Like none of this is a good idea. If you're using ChatGPT code for anything other than some kiddie scripts or quick problem solvers, you better know what you're getting.
Well, that’s pretty impressive
It usually takes me several hours to add two or three bugs to the one I’m looking for.