9 ms·

ChatGPT in Academic Writing

62 points by WolfOliver 4y ago

ryzvonusef 4y ago
see here is the thing, you didn't need Chat GPT to cheat. Nature's neural networks (also called humans) have been willing to provide cheating services for far cheaper than it costs OpenAI to run ChatGPT. So the question is, what were you doing then? ---- I'll share a personal anecdote on cheating and the futility of it. I am taking a course on SQL at Coursera, and at the end of the course there is a peer-reviewed project you have to do (you submit your project for others to mark, and you have to mark the project of four other students to pass). The thing is, you can already find the entire project online, a few people have posted their project for people to see, probably to help build their profile (just google "yelp coursera"). And obviously what most students did was just copy paste the damn thing, I immediately recognized that when I was marking the other candidates' projects, they didn't even bother attempting to hide it. The thing is, I am not doing this course because I am being forced to by work or anything, but because I wanted to learn. Therefore, copying pasting the project would have not achieved the goals. See, I wouldn't even have needed google to cheat, my BiL is a computer scientist, I could have easily used him to cheat if that was the goal. But I wanted to learn, so I did it on my own, as did many other students whose project I did end up agreeing to mark (I refused to mark the cheater's projects and just refreshed the tab) And since I am on HN and in tune with the tech on goings, I already had applied for the new BingGPT Chat, so I used to that to help with my project. ---- I didn't just paste my project questions, but I used it like I would have used my BiL to help; I asked it questions when my code didn't run like it should have. In other words, I would, say, google (and yes google, regular bing search is still bad) for questions like normal, but the answer from StackOverflow would be for, say, MySQL, so I would ask the BingGPT Chat to give me an example of doing the same thing in SQLite, which is what the course used. It would then recognize that I couldn't use CTE, for example (limitation of the Coursera web portal, not SQLite, in this instance), and give me an alternate way to do the same thing, and now I could use the concept to answer my own question. ---- In other words, ChatGPT is now acting as that education sibling/relative you can ask questions of to do homework... just people always have been before the existence of computers. The only benefit is that now EVERYONE has access to that educated "relative", not just the lucky ones. Now everyone can get a customized answer, and that's wonderful, imho. More importantly, is that you can now ask it follow up questions. Don't blame ChatGPT for cheating, people who wanted to cheat always could. ---- The problem with citing ChatGPT is, has anybody ever cited their relative who helped with the project? or do you cite google the search string you used? I understand what the author is getting at, but it's still strange. ChatGPT isn't the authoritative source, so why cite the prompt?
nestorD 4y ago
I am very tempted to try writing my next paper (maybe a low stake arxiv publication) with ChatGPT. Having done the research and gotten the results, asking it to turn draft paragraphs into properly written text.
- j45 4y ago
  There’s probably better writing tools to help with that than gpt.
  - sebzim4500 4y ago
    Like what? Other than hiring a copy editor, of course.
- corobo 4y ago
  You've done all the work only to be caught out by automated AI recognition[1] later. Low stakes should probably be fine but to anyone using it for something important at least do the old rewrite-to-not-be-plagiarism trick. I have to imagine there's going to be a lot of people regretting using GPT shortcuts soon. It'll be the equivalent of copy pasting Wikipedia by the end of the year I bet. The next big Google search ranking update should be fun too with all the AI content mills popping up right now. If you see your rival using AI for their blog posts, start writing real content in preparation :D [1] https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text https://openai.com/blog/new-ai-classifier-for-indicating-ai-...
  - Der_Einzige 4y ago
    These watermarking/"detect if someone is AI" techniques are trash, and OpenAI even admits it on their post about this. No proposed or implemented technique outside of this is even close to robust, reliable, or accurate. It is trivial to get around even the best of these techniques. This is a cat and mouse game with a fat, lazy cat and really motivated, olympic tier mice.
    - corobo 4y ago
jpwagner 4y ago
At some point the bibliography needs to reference the LLM itself which would need to be hosted indefinitely.
- WolfOliver 4y ago
  I do not think it is feasible to make sure prompts are reproducible. Considering that a LLMs are large you can not host every version of the model indefinitely.
  - jpwagner 4y ago
    if it's not reproducible, it's not science
    - ravi-delia 4y ago
      Some would argue that, due to the unfortunately near-universal deprecation of paper authors after 70-90 years, the actual process of writing any particular paper is not and has never been reproducible. As opposed to experiments, which are reproducible and are not generally contained within the operating weights of a LLM nor the thoughts of a human.
      - jpwagner 4y ago
        Your observation, while accurate, is a tangent. The point of the bibliography in the context of an academic paper is to reference the academic merit of the work. In the case of science, this would be reproducible experiments (ideally). Perhaps you would prefer to include the generated text source as an author.
        ravi-delia 4y ago
        I'd rather not include it at all, for exactly that reason. It's just a writing tool- the paper is either correct or incorrect on the same basis as any other paper. We include bibliographies to ensure that the relevant scientific data is present, but I don't think there's any reason to say that a non-reproducible abstract isn't science
neom 4y ago
My wife is a university professor in the states, she said it's incredibly obvious who has started to use a GPT for the work they turn in. There is a lot of wrestling going on within the staff in her department on what they should do, some people want to outright ban anything that "seems" it was written by a LLM. As my wife discussed with me, it's hard to know how much was written by GPT and how much was edited by the student, maybe they asked 50 questions to the LLM and then copy/pasted it together themselves, maybe they used it for the outline, and filled in the details themselves. The lines can get pretty blurry. One thing that it seems most of the profs find annoying is... they're aware that GPT can invent convincing sounding sources, so they're doing a lot more work verifying sources. Commented this ^^ the other day on the UK students being able to use LLMs post: https://news.ycombinator.com/item?id=35028224 https://news.ycombinator.com/item?id=35028224
- euroderf 4y ago
  Students should be rewarded for checking sources, and (moderately?) penalized for not doing so. I would say that at a minimum, every fictitious source knocks off a full letter grade. And make sure students know this. They will adapt. And it might get them to the sources, at least superficially (yeay?) to check for existence/non-existence.
- hypertele-Xii 4y ago
  Interview the student. That's literally all it takes... but oh no, it's too expensive, must figure out how to automate testing... while fighting against the automation of answering.
- gimili 4y ago
  I found it interesting that there start to be best-practices for professors and teachers on how to deal with ChatGPT in a class room [1]. One I really liked a lot: >And if you don't trust the AI to output correct content, teach your students critical thinking by using Assisted Teaching to write history essays and let your students find mistakes in it. Another way (if you want to go the more traditional way), is the Flipped Classroom [2] method, in which students self-teach at home and do the "homework" in the classroom. I agree with others here that there is no way to ban these technologies; there are just better and worse ways to deal with it. [1]: https://assistedeverything.substack.com/p/the-age-of-assisted-everything https://assistedeverything.substack.com/p/the-age-of-assiste... [2]: https://en.wikipedia.org/wiki/Flipped_classroom https://en.wikipedia.org/wiki/Flipped_classroom
bee_rider 4y ago
Avoiding doing assignments by having chatGPT do them is obviously a bad idea. Cheating has always been possible, it always will be, we try to prevent it, but cheaters are really creative and mostly hurting themselves. Any effort spent on preventing students from ripping themselves off is effort not spent on helping the students who are engaging with the system in good faith and trying to learn. It seems more troubling if people can cheat their way all the way through college. There are a number of skills that really benefit from kind of boring/non-creative work. In intro classes, part of it is just spending enough time learn “reflexes” that will help later. I’m sure it is easy to find tools that will do this work for you. But at some point, I mean, junior and senior (Err, I guess this is US lingo maybe, last couple years of college) should be doing actually interesting work, to the point where if they can “cheat” their way through the material with new tools, then… they’ve found an interesting new way to solve real problems. And if they aren’t doing anything interesting at all, then there are much, much deeper problems at the school.
- Balgair 4y ago
  I don't know how many HNers were in greek life, but every single house has a huge test and homework bank. It's one of the big draws for pledging.
- WalterBright 4y ago
  > learn “reflexes” that will help later True enough. I inadvertently learned trig so thoroughly that I could "see" many steps ahead as one. I say inadvertently because since every class was a math class, full of incidental trig work, this winds up happening to everyone.
- bitL 4y ago
  This only works until people notice that many at the top (company/society) were the cheaters.
  - runsWphotons 4y ago
    honestly doing my assignments seems to have gotten many of my less honest peers better careers. (my smarter peers mostly did better too though)
yawnxyz 4y ago
ChatGPT has been great at creating summaries for me, even for pieces that I've written. I use ChatGPT to riff off ideas and make sure I haven't missed anything; often ChatGPT reminds me that "oh right I'm missing aspect X Y and Z which would make my project or writing more complete. With that said, I haven't written anything published in a journal yet, and maybe I'll use ChatGPT to assist (probably in the same way as I mentioned, as a way to riff off and brainstorm ideas, rather than as a content generator; its output is way too boring and verbose for my linking)
- sebzim4500 4y ago
  >its output is way too boring and verbose for my linking Have you tried adjusting your prompts? Even just a single instruction to 'be concise' will make a difference. For some applications, you might be able to give examples fo the writing style you are looking for.
- cbozeman 4y ago
  I'm extremely verbose and proper in written communication by nature, so none of my clients have even noticed that I've been using ChatGPT in my workflow for the past three weeks. Personally I couldn't be more tickled shitless at this creation. It's cheap ($20 a month!!), it's fast, it's already learned my writing and coding style, and it is probably saving me 30-40% time every week. The amount of projects I've been able to start working on, that I had on the back burner, is just incredible. I really could not be more excited about where GPT 4.0 and 5.0 will go.
dalmo3 4y ago
Huge business opportunity for OpenAI in selling GPT detection to teaching institutions? Give away the poison, sell the antidote.
- WolfOliver 4y ago
  https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text https://openai.com/blog/new-ai-classifier-for-indicating-ai-...
armchairhacker 4y ago
I thought this would talk about ChatGPT in academic research writing i.e. papers. Many academics are really smart but not very good at explaining things, especially when using a non-native language. Many papers don't get great ideas don't get those ideas the reach they deserve simply because they are too dense to understand, not just in the content itself, but in the overly-verbose and convoluted writing. If ChatGPT helps academics write better papers it would be amazing. And presumably these papers' logic is actually checked by people (even if they don't go through peer-review) so potential BS is not a problem.
imchillyb 4y ago
Academia needs its own version of ChatGPT that parses through the student's assignments, and asks the student a series of AI generated questions based on their own work. The question and answer session would be done through a video session and recorded. Seems fairly simple. Answer the question. 10.... 9.... 8.... 7... Students who wrote their own assignments would be able to answer the questions and those who didn't, couldn't.

62 points by WolfOliver 4y ago

see here is the thing, you didn't need Chat GPT to cheat. Nature's neural networks (also called humans) have been willing to provide cheating services for far cheaper than it costs OpenAI to run ChatGPT. So the question is, what were you doing then? ---- I'll share a personal anecdote on cheating and the futility of it. I am taking a course on SQL at Coursera, and at the end of the course there is a peer-reviewed project you have to do (you submit your project for others to mark, and you have to mark the project of four other students to pass). The thing is, you can already find the entire project online, a few people have posted their project for people to see, probably to help build their profile (just google "yelp coursera"). And obviously what most students did was just copy paste the damn thing, I immediately recognized that when I was marking the other candidates' projects, they didn't even bother attempting to hide it. The thing is, I am not doing this course because I am being forced to by work or anything, but because I wanted to learn. Therefore, copying pasting the project would have not achieved the goals. See, I wouldn't even have needed google to cheat, my BiL is a computer scientist, I could have easily used him to cheat if that was the goal. But I wanted to learn, so I did it on my own, as did many other students whose project I did end up agreeing to mark (I refused to mark the cheater's projects and just refreshed the tab) And since I am on HN and in tune with the tech on goings, I already had applied for the new BingGPT Chat, so I used to that to help with my project. ---- I didn't just paste my project questions, but I used it like I would have used my BiL to help; I asked it questions when my code didn't run like it should have. In other words, I would, say, google (and yes google, regular bing search is still bad) for questions like normal, but the answer from StackOverflow would be for, say, MySQL, so I would ask the BingGPT Chat to give me an example of doing the same thing in SQLite, which is what the course used. It would then recognize that I couldn't use CTE, for example (limitation of the Coursera web portal, not SQLite, in this instance), and give me an alternate way to do the same thing, and now I could use the concept to answer my own question. ---- In other words, ChatGPT is now acting as that education sibling/relative you can ask questions of to do homework... just people always have been before the existence of computers. The only benefit is that now EVERYONE has access to that educated "relative", not just the lucky ones. Now everyone can get a customized answer, and that's wonderful, imho. More importantly, is that you can now ask it follow up questions. Don't blame ChatGPT for cheating, people who wanted to cheat always could. ---- The problem with citing ChatGPT is, has anybody ever cited their relative who helped with the project? or do you cite google the search string you used? I understand what the author is getting at, but it's still strange. ChatGPT isn't the authoritative source, so why cite the prompt?

nestorD 4y ago

I am very tempted to try writing my next paper (maybe a low stake arxiv publication) with ChatGPT. Having done the research and gotten the results, asking it to turn draft paragraphs into properly written text.

j45 4y ago
There’s probably better writing tools to help with that than gpt.
- sebzim4500 4y ago
  Like what? Other than hiring a copy editor, of course.
corobo 4y ago
You've done all the work only to be caught out by automated AI recognition[1] later. Low stakes should probably be fine but to anyone using it for something important at least do the old rewrite-to-not-be-plagiarism trick. I have to imagine there's going to be a lot of people regretting using GPT shortcuts soon. It'll be the equivalent of copy pasting Wikipedia by the end of the year I bet. The next big Google search ranking update should be fun too with all the AI content mills popping up right now. If you see your rival using AI for their blog posts, start writing real content in preparation :D [1] https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text https://openai.com/blog/new-ai-classifier-for-indicating-ai-...
- Der_Einzige 4y ago
  These watermarking/"detect if someone is AI" techniques are trash, and OpenAI even admits it on their post about this. No proposed or implemented technique outside of this is even close to robust, reliable, or accurate. It is trivial to get around even the best of these techniques. This is a cat and mouse game with a fat, lazy cat and really motivated, olympic tier mice.
  - corobo 4y ago

jpwagner 4y ago

At some point the bibliography needs to reference the LLM itself which would need to be hosted indefinitely.

WolfOliver 4y ago
I do not think it is feasible to make sure prompts are reproducible. Considering that a LLMs are large you can not host every version of the model indefinitely.
- jpwagner 4y ago
  if it's not reproducible, it's not science
  - ravi-delia 4y ago
    Some would argue that, due to the unfortunately near-universal deprecation of paper authors after 70-90 years, the actual process of writing any particular paper is not and has never been reproducible. As opposed to experiments, which are reproducible and are not generally contained within the operating weights of a LLM nor the thoughts of a human.
    - jpwagner 4y ago
      Your observation, while accurate, is a tangent. The point of the bibliography in the context of an academic paper is to reference the academic merit of the work. In the case of science, this would be reproducible experiments (ideally). Perhaps you would prefer to include the generated text source as an author.
      - ravi-delia 4y ago
        I'd rather not include it at all, for exactly that reason. It's just a writing tool- the paper is either correct or incorrect on the same basis as any other paper. We include bibliographies to ensure that the relevant scientific data is present, but I don't think there's any reason to say that a non-reproducible abstract isn't science

neom 4y ago

My wife is a university professor in the states, she said it's incredibly obvious who has started to use a GPT for the work they turn in. There is a lot of wrestling going on within the staff in her department on what they should do, some people want to outright ban anything that "seems" it was written by a LLM. As my wife discussed with me, it's hard to know how much was written by GPT and how much was edited by the student, maybe they asked 50 questions to the LLM and then copy/pasted it together themselves, maybe they used it for the outline, and filled in the details themselves. The lines can get pretty blurry. One thing that it seems most of the profs find annoying is... they're aware that GPT can invent convincing sounding sources, so they're doing a lot more work verifying sources. Commented this ^^ the other day on the UK students being able to use LLMs post: https://news.ycombinator.com/item?id=35028224 https://news.ycombinator.com/item?id=35028224

euroderf 4y ago
Students should be rewarded for checking sources, and (moderately?) penalized for not doing so. I would say that at a minimum, every fictitious source knocks off a full letter grade. And make sure students know this. They will adapt. And it might get them to the sources, at least superficially (yeay?) to check for existence/non-existence.
hypertele-Xii 4y ago
Interview the student. That's literally all it takes... but oh no, it's too expensive, must figure out how to automate testing... while fighting against the automation of answering.
gimili 4y ago
I found it interesting that there start to be best-practices for professors and teachers on how to deal with ChatGPT in a class room [1]. One I really liked a lot: >And if you don't trust the AI to output correct content, teach your students critical thinking by using Assisted Teaching to write history essays and let your students find mistakes in it. Another way (if you want to go the more traditional way), is the Flipped Classroom [2] method, in which students self-teach at home and do the "homework" in the classroom. I agree with others here that there is no way to ban these technologies; there are just better and worse ways to deal with it. [1]: https://assistedeverything.substack.com/p/the-age-of-assisted-everything https://assistedeverything.substack.com/p/the-age-of-assiste... [2]: https://en.wikipedia.org/wiki/Flipped_classroom https://en.wikipedia.org/wiki/Flipped_classroom

bee_rider 4y ago

Avoiding doing assignments by having chatGPT do them is obviously a bad idea. Cheating has always been possible, it always will be, we try to prevent it, but cheaters are really creative and mostly hurting themselves. Any effort spent on preventing students from ripping themselves off is effort not spent on helping the students who are engaging with the system in good faith and trying to learn. It seems more troubling if people can cheat their way all the way through college. There are a number of skills that really benefit from kind of boring/non-creative work. In intro classes, part of it is just spending enough time learn “reflexes” that will help later. I’m sure it is easy to find tools that will do this work for you. But at some point, I mean, junior and senior (Err, I guess this is US lingo maybe, last couple years of college) should be doing actually interesting work, to the point where if they can “cheat” their way through the material with new tools, then… they’ve found an interesting new way to solve real problems. And if they aren’t doing anything interesting at all, then there are much, much deeper problems at the school.

Balgair 4y ago
I don't know how many HNers were in greek life, but every single house has a huge test and homework bank. It's one of the big draws for pledging.
WalterBright 4y ago
> learn “reflexes” that will help later True enough. I inadvertently learned trig so thoroughly that I could "see" many steps ahead as one. I say inadvertently because since every class was a math class, full of incidental trig work, this winds up happening to everyone.
bitL 4y ago
This only works until people notice that many at the top (company/society) were the cheaters.
- runsWphotons 4y ago
  honestly doing my assignments seems to have gotten many of my less honest peers better careers. (my smarter peers mostly did better too though)

yawnxyz 4y ago

ChatGPT has been great at creating summaries for me, even for pieces that I've written. I use ChatGPT to riff off ideas and make sure I haven't missed anything; often ChatGPT reminds me that "oh right I'm missing aspect X Y and Z which would make my project or writing more complete. With that said, I haven't written anything published in a journal yet, and maybe I'll use ChatGPT to assist (probably in the same way as I mentioned, as a way to riff off and brainstorm ideas, rather than as a content generator; its output is way too boring and verbose for my linking)

sebzim4500 4y ago
>its output is way too boring and verbose for my linking Have you tried adjusting your prompts? Even just a single instruction to 'be concise' will make a difference. For some applications, you might be able to give examples fo the writing style you are looking for.
cbozeman 4y ago
I'm extremely verbose and proper in written communication by nature, so none of my clients have even noticed that I've been using ChatGPT in my workflow for the past three weeks. Personally I couldn't be more tickled shitless at this creation. It's cheap ($20 a month!!), it's fast, it's already learned my writing and coding style, and it is probably saving me 30-40% time every week. The amount of projects I've been able to start working on, that I had on the back burner, is just incredible. I really could not be more excited about where GPT 4.0 and 5.0 will go.

dalmo3 4y ago

Huge business opportunity for OpenAI in selling GPT detection to teaching institutions? Give away the poison, sell the antidote.

WolfOliver 4y ago
https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text https://openai.com/blog/new-ai-classifier-for-indicating-ai-...

armchairhacker 4y ago

I thought this would talk about ChatGPT in academic research writing i.e. papers. Many academics are really smart but not very good at explaining things, especially when using a non-native language. Many papers don't get great ideas don't get those ideas the reach they deserve simply because they are too dense to understand, not just in the content itself, but in the overly-verbose and convoluted writing. If ChatGPT helps academics write better papers it would be amazing. And presumably these papers' logic is actually checked by people (even if they don't go through peer-review) so potential BS is not a problem.

imchillyb 4y ago

Academia needs its own version of ChatGPT that parses through the student's assignments, and asks the student a series of AI generated questions based on their own work. The question and answer session would be done through a video session and recorded. Seems fairly simple. Answer the question. 10.... 9.... 8.... 7... Students who wrote their own assignments would be able to answer the questions and those who didn't, couldn't.