Is it okay to use ChatGPT for proofreading?
I sometimes use chatGPT to proofread longer texts (like 1000+ words) I write in English. Although this is not my first language, I often find myself writing in English even outside of internet forums. That is because if I read or watch something in English, and that thing motivates me to write, my brain organically gravitates toward it.
My English is pretty good and I am reasonably confident communicating in that language, but it will never be the same as my native language. So I will often run my stuff through Grammarly and chatGPT. If you wanna say "This will teach you bad habits", please don't. Things like Grammarly and Google Translate taught me so much and improved my English so much, that I am a bit tired of that line of reasoning. I read most of my books in English. I'm not a beginner so I can and do check for all the changes, and vet them myself as I don't always agree with them.
With GPT, I usually just ask it to elaborate a critique rather than spit out a corrected version. Truth be told, when I did ask for a corrected version, it made plenty of sensible corrections that didn't really alter anything other than that. So I guess I just wanna know everyone's feelings about this. Suppose I write a bunch, have GPT correct it for me, compare it with the original and verify every correction. Is that something you would look at unfavorably?
Thanks!
Ever since ChatGPT came out, I have been saying that this is the perfect use case for it. It is importantly that you don’t plagiarize the output, but it sounds like you don’t. I’m a native English speaker, and I don’t think it will teach you bad habits. It might if you just have ChatGPT rewrite things for you. But if you use it to give you advice and rewrite it yourself, that is fine in my opinion.
Not at all. I’ve been really negative about the whole AI hype and hate the useless things it has been added to, but this definitely isn’t one of them. This sounds like a perfect use case for improving written text and leveraging the ability of LLMs.
Only thing I’m wondering is if ChatGPT is the most convenient/best fit for this. Maybe there are less evil alternatives that provide the same functionality?
Why is chatGPT evil?
It's a live service provided by what (is? soon will be?) a for-profit (non-B?) corporation, which necessarily means it is designed to put the corporation's profits over the interests of the user.
In this case that might look like an incentive to cultivate a dependence and then jack up the price, a lack of attention (or of drawing the user's attention) to problems with the output that are cheaper to pay damages over than to fix, or a standard enshittification where it finds people like advertisers or ad-targeting companies to sell the user out to.
There's also the problem where people doubt they have clear title to their training data, but they use it anyway. And the way they jealously guard their position and hipocritically get mad every time somebody like maybe DeepSeek trains on data generated from their model.
Personally, no. It’s a tool and should be used like one. Some of this AI pushback vaguely reminds me of people hating on spell check for making them worse at spelling, and yet it’s widely adopted now.
I think the big issue is when you let it think for you or when it lies to you/changes the meaning in ways you don’t understand or don’t notice.
Same with the pushback people had on Wikipedia when it was launched.
Sounds like a great use of AI to me, I have it do this for me with Spanish on the rare occasion that I still need to write Spanish (first pass is me + searching WordReference, then I have chatgpt correct it)
FYI Grammarly pro is a total ripoff, most of the suggestions that are in pro but not free are.........kinda bad/wrong. At least that was the case when I used it ~2 years ago
Oh, I don't use Grammarly Pro. I wouldn't pay for it. It is way too intrusive. It sometimes gives me free PRO hints and I wish I could disable those.
Yeah that's the correct decision. I wouldn't even be tempted by a free trial, it's legitimately a worse product than the free version imo
(which it kinda has to be, because if the free version doesn't totally fix all your black-and-white mistakes, no one will use it; so the only features available to include in pro are alternatives that may or may not be better, and it doesn't know mood enough to be remotely accurate on those)
Grammarly pro also frankly seems quite expensive compared to competitors, namely Antidote and LanguageTool
The former I hear about a lot more in my day to day IRL life, and is seemingly everywhere in personal and professional life in the part of Canada I'm in. The latter I see mentioned much more online especially since it integrates natively with LibreOffice.
I am going to come in with a rare no: do not use ChatGPT for proofreading.
You should be able to do LLM-based proofreading with a much lighter, cheaper, and more efficient tool than the umpteen-billion-parameter, needs-more-GPUs-than-you-can-buy ChatGPT. Something that takes a tiny model and runs it over a few dozen tokens of context and lights up red or something when the text is highly improbable. You ought to be able to solve the problem pretty well with a few watts on a laptop, so the extra hardware and energy used by doing it with millitary-grade ChatGPT aren't justified and thus should not be used.
Here's an example of local grammar checking on Mac with a 7B parameter model, with a bunch of Automator stuff to tie in to text boxes in apps, though I think it still uses Q&A and not the model's probabilities directly, which probably would be better.
I should clarify that I never paid for GPT and do not intend to do so. I have considered running something like that locally but in my impression, it doesn't seem very easy to do so. I am neither a programmer, I am not in IT, and I am not in STEM at all. I write on an extremely weak laptop with very little RAM, a weak CPU and no dedicated GPU. It was, quite literally, the cheapest and slowest laptop I could find that was not a toy :P
Yeah, one thing ChatGPT has going for it is that it is in fact a product and not a pile of script glue. And you might not have the local power to run a model as big as you need.
But it still might be worth looking for less-powerful options, or options that don't grant OpenAI so many rights to your text.
I don't think it's a bad thing to use LLMs that way at all, as long as you keep in mind that it's at its very core a token-guessing-machine and treat the input as such. The more you use it, the higher the chance is that it'll say something completely wrong.
I've used it for cover letters in the past and the result was mixed. Sometimes it gave a useful point but often it was a vague compliment while telling me to be more specific about one or two things when it already was as specific as possible without making it a huge story.
Yes, I find GPT terrible for criticism of content. For me, it only works for basic proofreading.
For grammar and spelling I'd say that is definitely okay. For sentence structures and punctuation I would remain critical of what chatGPT is suggesting. You mentioned you already do that with tools like grammarly, so it probably is okay. But, I wanted to mention it anyway to be sure.
As far as my reasoning goes, LLMs tend to stick to specific writing patterns. Some of those patterns aren't that common in regular English use. It is part of what makes chatGPT output at times recognizable for people.
nah I'd trust chatgpt over grammarly for sentence structure any day of the week
grammarly is honestly just bad for anything except basic rules
(yes, I have a deep (and rational) hatred of grammarly)
That's a false equivalence though. Sure chatGPT might be better, but it is not perfect either.
oh yeah I'm not saying it's perfect. But IME it's better than grammarly
Putting aside my grievances with OpenAI's catastrophic failures on the ethical front and therefore ChatGPT by extension, which, while tangentially relevant, I have already gone at length about, I do think it makes sense to use an LLM to help with proofreading, provided you have the experience required to recognize what is and isn't sensible output from the LLM. I wouldn't be comfortable with suggesting to someone who isn't confident regarding how fluent they are in English to rely on LLM generated input since a very relevant limitation of LLMs is giving confidently wrong answers, but since you're already primed by definition to double-check everything when you're proofreading, it could meaningfully contribute to speeding up the process. It's worth trying out, at least.
If you want to use ChatGPT as a tool to write better, you can also use it as a tool to identify specific areas of improvement. You can give it your first drafts and final versions that you ended up going with after the critique or slight corrections from your previous chats. Ask for what patterns it sees in the first drafts and the changes made, and if there are any suggestions it has for avoiding repeating the same kind of errors.
Another great free tool is Hemingway, which is especially useful if you're aiming for clarity since it will point out any complicated sentence structures, filler words, and give you a readability grade for your text. They've added paid/AI features, but the core functionality is something I've been using for years alongside Grammarly (also the free version).
I have indeed used the Hemingway website. It is aptly named because the style it suggests is very much in tune with my experience reading Ernest Hemingway. The Hemingway App really wants me to write very short sentences, and Hemingway is to adjectives and adverbs what an oncologist is to cancer. It is too much sometimes. But it is a good source of critique if you don't obsess about it.
Yeah, I find it handy for suggestions, just in case I might want to reconsider writing a sentence differently. But like with any other by-the-rules writing feedback, it's best to take what you like. Writing in your voice should be about what feels right to you and being understood.
A lot of people have already shared a lot of information on ChatGPT / LLMs, but I do feel like there is one important thing regardless of what you choose.
Regardless of what solution you choose, you definitely need to be able to tell when a grammar correction tool gives incorrect feedback. Because of this a big criteria for me is that the tool in question can explain why it's suggesting a change. Whether its a specific grammar rule being violated, register of language etc.
Amazed that I had to scroll at all:
Yes.
This comment will just become noise the moment you address it, but I did feel the need to leave it as a heads up:
(folded to avoid meta clutter)
The thread's title (Is it okay to use ChatGPT for proofreading?) and the last question of the post's body (Is that something you would look at unfavorably?) makes answering "yes" ambiguous. I can't tell whether you meant "Yes, it is okay to use chatgpt for proofreading." answering the title (which intuitively makes sense) or "Yes, that is something I would look at unfavorably" answering the post's last question (as this reply did, starting with "Not at all" but actually answering the title's question in the affirmative, and this one which answered "no" to using ChatGPT specifically but is favorable to LLMs in general) which would completely flip your answer.
Unless it's neither and you actually meant to answer "Yes" to both questions, considering using ChatGPT to be okay but at the same time still something you would look at unfavorably, and while I can see that being a coherent stance that would be an even more unlikely interpretation from just a one word "Yes" answer and I'm definitely overthinking it at this point.
Ha! Well, being mindful that I don't log on to the internet to start arguments ... I meant to make a pithy reply in my original comment by simply saying, "yeah, it's fine to use LLMs to proof-read." -- and it was intentionally trying to zero-out the distinction between taking some of the suggestions and utilizing them, versus just reviewing what the LLM might be suggesting.
To be completely clear (but less pithy, which was the goal!):
Now, if we want to dive into the more philosophical aspects, I'd be game. But the reason I provided the original pithy one word reply was just because of the way the question was phrased/setup - as a binary yes/no is this OK or not. It wasn't really an invitation (including in the body) to have a deeper philosophical conversation!
:)
Fair enough, that was what I figured was the most likely meaning but the surrounding context
and my tendency to hopelessly overthink everythingcast enough doubt in my mind that I wasn't actually sure, and given that a credible way of interpreting your reply meant the complete opposite of your actual answer, I felt the need to ask. Thank you for clearing it up.