Timeline for Examples of ChatGPT being right and wrong on Stack Exchange?
Current License: CC BY-SA 4.0
58 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 10, 2023 at 19:01 | comment | added | ggorlen | @Braiam By the way, if my first response isn't satisfying for you linguistically, you may read "determining value" as "correctness" if you want. If the answers were so correct and insightful, OPs would have gladly accepted them and upvoted them. But they didn't in almost every case. Why not? Because they were wrong. I'm essentially answering your question by letting acceptance and upvotes represent a good heuristic to determine "correctness", since I can't sit down and mathematically, incontrovertibly prove correctness (or whatever you seem to be expecting when you say "correctness"). | |
| Jul 10, 2023 at 18:52 | comment | added | ggorlen | @Braiam If you want, I can spend a couple hours dredging up 10 answers and labeling them correct or incorrect, but you'll find a way to dismiss that too, so it's pointless. Let's just drop it--you won't be convinced by whatever results I show. Anyone with your level of experience who can't see for themselves what an unmitigated trainwreck LLMs have been for spam answering on SO and quality control in general is beyond my ability to convince in a limited time, and I've already invested a good 10 or 12 hours into this post. Maybe things are different over at Ubuntu and U&L, but I doubt it. | |
| Jul 10, 2023 at 18:44 | comment | added | ggorlen | @Braiam Please revisit the last sentence of my answer, which labels two answers as correct and incorrect, to help assuage the pedantry you seem to be using as a tool to ignore the clear results I'm showing. | |
| Jul 10, 2023 at 18:37 | comment | added | ggorlen | @Braiam I don't know anything about Chinese, but Rebecca has only shown 9 answers and 2/9 were "rubbish"! That's horrible accuracy. | |
| Jul 10, 2023 at 18:31 | comment | added | ggorlen | @Braiam We're going in circles, just like most of my interactions with ChatGPT. I showed a list of all 114 suspected GPT answers I've flagged on SO, and I showed that although these answers had been active for an average 23 hours before flagging, almost none of them turned out to be upvoted or accepted, the usual metrics we use for determining value on Stack Overflow. Their aggregate value was overwhelmingly negative, in spite of the few that appear to be correct, which I identified as I can. | |
| Jul 10, 2023 at 18:27 | comment | added | Braiam | correctness given the query. You did the opposite, you posted answers and claimed that they aren't "useful" without the evaluation if they are correct or not. Unless you can list answers that are posted on SE, and you can offer an evaluation if they are correct or wrong given the question, you aren't answering my question, neither the title nor the body of it. | |
| Jul 10, 2023 at 18:25 | comment | added | Braiam | @ggorlen as the Hitchens's razor goes: what can be asserted without evidence can also be dismissed without evidence. I'm not rejecting that you bring evidence, I'm rejecting that you do not bring the evidence we need to answer a question: "are answers posted on SE generated by LLM more likely to be right or wrong?" This is the 4th time in this comment thread that I have asserted this. I do not know how else I can assert it. Rebecca J. Stones understood the query I'm making, that's why didn't post an answer, since those answers/examples didn't exist on SE but offer an evaluation of.... | |
| Jul 2, 2023 at 14:34 | history | edited | ggorlen | CC BY-SA 4.0 |
adjustment
|
| Jul 1, 2023 at 15:52 | comment | added | ggorlen | @Braiam It's funny you want to "stop the rhetoric and bring facts", yet you've done nothing but present rhetoric without facts in this discussion, while slamming my post that brings way more data to the debate than the answers you support which are clearly pro-LLM cherry-pickers. If my answer seems biased, it's because GPT has had such an overwhelmingly negative impact on the site. Like I said, I've tried to present all of the positive LLM stuff I've seen on the site, but there's not much of it that's obvious to me. It's largely zero-effort, rep farming spam. | |
| Jul 1, 2023 at 15:50 | comment | added | ggorlen | @Braiam Unfortunately, I don't have any way to prove an exact percentage of GPT correctness, but the ballpark estimate depicted here is not looking good. Of the posts I flagged, I'd guess maybe 10% are "correct". How do you define "correct", anyway? If it's defined as the OP accepting the answer, then less than 1% are correct among my flags. In many cases, OP has responded to beg people to stop using LLMs. As I've argued, you're asking the wrong question--it's an XY problem. Even if 100% were correct (and we're far from that), it's still not worth turning SO into a dumping ground of LLM spew | |
| Jul 1, 2023 at 15:45 | comment | added | ggorlen | @Braiam I've given you all of the data I personally have access to in my experience dealing with LLMs on the site. I'm just repeating myself here, but it's clear from this post that there's little to no value, in aggregate, of the GPT spam I've personally seen and flagged. I explain my flagging methodology and cite cases where the answer is correct and cases where the answer is incorrect. Choosing one or the other in isolation, as you and other answers in this thread have done, is cherry-picking. | |
| Jul 1, 2023 at 11:56 | comment | added | Braiam | @ggorlen "What we're interested in is the impact on the site, right?" is that what my title/body of my question says? No. That's not what this question is about. Stop trying to derail the question with topics discussed elsewhere. I'm looking for data. Data of ChatGPT being right or wrong on the site. One of the argument to ban the use on the site is that, and I quote "Overall, because the average rate of getting correct answers from ChatGPT and other generative AI technologies is too low", but how low? 50%, 25%, 95%. That's what this question is about, stopping the rhetoric and bringing facts. | |
| Jun 26, 2023 at 3:38 | history | edited | ggorlen | CC BY-SA 4.0 |
clarify exact time between posting and flagging
|
| Jun 21, 2023 at 20:39 | history | edited | ggorlen | CC BY-SA 4.0 |
clarify
|
| Jun 21, 2023 at 20:12 | history | edited | ggorlen | CC BY-SA 4.0 |
add flag proof
|
| Jun 21, 2023 at 19:56 | history | edited | ggorlen | CC BY-SA 4.0 |
add links to correct and incorrect posts
|
| Jun 21, 2023 at 19:51 | history | edited | ggorlen | CC BY-SA 4.0 |
add links to correct and incorrect posts
|
| Jun 21, 2023 at 19:13 | history | edited | ggorlen | CC BY-SA 4.0 |
add link to identify LLM answer
|
| Jun 21, 2023 at 18:35 | history | edited | ggorlen | CC BY-SA 4.0 |
remove self-delete
|
| Jun 21, 2023 at 18:24 | history | edited | ggorlen | CC BY-SA 4.0 |
add bolding
|
| Jun 21, 2023 at 17:59 | history | edited | ggorlen | CC BY-SA 4.0 |
add total statistics
|
| Jun 21, 2023 at 17:54 | history | edited | ggorlen | CC BY-SA 4.0 |
add total statistics
|
| Jun 21, 2023 at 17:45 | history | edited | ggorlen | CC BY-SA 4.0 |
add total statistics
|
| Jun 21, 2023 at 17:38 | history | edited | ggorlen | CC BY-SA 4.0 |
add total statistics
|
| Jun 21, 2023 at 15:46 | comment | added | ggorlen | @Braiam Anyway, looks like you've had this discussion before and I don't think you'll care about the evidence I'm presenting here that clearly disproves your thesis. "I've not seen examples myself of answers of ChatGPT being flat out wrong" -- now you have. Or just ask it any 10 non-trivial questions yourself. | |
| Jun 21, 2023 at 15:29 | comment | added | ggorlen | @Braiam In any case, a question like "Examples of ChatGPT being right and wrong on Stack Exchange?" totally misses the point and is an XY problem, as I've shown, which is why it's so heavily downvoted. Disappointing to see a respected expert who I've learned a good deal from over the years "accept rubbish" in spite of their profile quote. | |
| Jun 21, 2023 at 15:27 | comment | added | ggorlen | @Braiam Sorry, but I don't have the ability or time to give you a precise % accuracy--that's too much to ask, but hopefully someone has the time, talent and dataset to cook it up. Anecdotally based on my usage and from what I've seen on SO, maybe 25% correct, which is way too low. Even if it's 99% though, I still don't see a reason to post LLM answers on SO. Just ask the LLM directly if you want an LLM answer--simple, everyone wins. LLM flooding all platforms is pointless and redundant at best, and at worst drowns out existing human knowledge. And that's at best, which we're nowhere near. | |
| Jun 21, 2023 at 15:18 | comment | added | ggorlen | @Braiam As mentioned, the users spamming these LLM answers don't even care if they're right or not. They don't possess the knowledge to evaluate LLM correctness. The accuracy is clearly dismal and the answers are clearly generally unhelpful, in spite of occasional lucky hits. If this wasn't the case, we'd be seeing upvotes on these answers and OPs enthusiastically thanking LLM spammers in the comments for saving them the trouble of typing the question into an LLM for them and deigning to copy-paste the answer for them, as if they hadn't thought of doing so themselves. There'd be no argument. | |
| Jun 21, 2023 at 15:13 | comment | added | ggorlen | @Braiam Yes, they've cherry picked a few correct answers. So what? We all know LLMs can give right answers. What we're interested in is the impact on the site, right? Those answers don't add any information to the conversation. It's much more compelling to simply grab the latest flag spammer and go from there. Did you check out my flags as I suggested? There are hundreds of others you can pick from if you think my process is biased. If you don't like my flags, look at flags from another user. The overall impact is clearly negative. What's your agenda with this question? | |
| Jun 21, 2023 at 14:05 | comment | added | Braiam | No cherry picked samples of users, but the answers themselves. Franck posted an answer where he shows 3 correct answers, I posted in my question 2 examples. Both users give a objective observation: these answers are right/wrong and how. | |
| Jun 21, 2023 at 14:03 | comment | added | Braiam | "most of the answers I cite are unhelpful" by being unhelpful yourself. What are we learning that we didn't know with your answer? Nothing. My question is asking for a particular datapoint that is mostly ignored in the community: the purported accuracy of generated answers. I haven't seen anyone that claims that generated answers are wrong, with evidence that objectively points that x% of the answers are wrong. My observations of these answers is 75/25 chatgpt generating something accurate. That's the purpose of this question, to test our anecdotes and biases with actual data. | |
| Jun 20, 2023 at 17:20 | comment | added | ggorlen | @Braiam I suspect people such as yourself are having such a rough time with an answer that disagrees with your point of view is that ChatGPT will rarely disagree with you. The conversation is completely artificial: "You're right. I apologize for my mistake in my last answer. Here's updated answer that doesn't challenge your viewpoint about the world whatsoever". ChatGPT seems fantastic for confirmation bias and will rarely let you know when you have an XY problem or push back on anything. | |
| Jun 20, 2023 at 17:14 | comment | added | ggorlen | @Braiam There's a huge difference between spamming LLM answers and writing a human answer as I've done here. This answers your question in the aggregate: most of the answers I cite are unhelpful, and I show a case of "ChatGPT being flat out wrong" as determined by OP: "I already had most of this suggested by ChatGPT anyway and no luck". The problem with showing singular examples is that, yes, ChatGPT is sometimes correct. Other answers here have cherry-picked a success story, while I've simply grabbed my most recent LLM encounter to show the actual impact of LLMs on the site. | |
| Jun 20, 2023 at 15:10 | comment | added | Braiam | "users like this are pulling random posts in random tags and copy-pasting ChatGPT output" and that give you the right to ignore my question which two other users correctly answer (giving examples). You do exactly what you complain these users are doing, just because. This answer is no different of what you accuse that user of doing. Crapshooting a bunch of things without regard to address the question itself. | |
| Jun 19, 2023 at 22:48 | comment | added | ggorlen | @Braiam Also, my point about not being a SME in all these tags is there to illustrate that, quite clearly, users like this are pulling random posts in random tags and copy-pasting ChatGPT output. The answerer here has no idea what they're talking about, could not answer these questions on their own without AI help, and could care less whether or not the ChatGPT output is correct! And this behavior is now endorsed! By and large, LLMs are not being used as an assistant to preexisting expertise as suggested by this well-intentioned answer. | |
| Jun 19, 2023 at 22:02 | comment | added | ggorlen | @Braiam If you don't believe me, here's my flags: stackoverflow.com/users/flag-summary/6243352. Scroll back from the start until you see a big chunk of red--that's what you see in this post here. Keep scrolling back and you'll see more, larger chunks of red. | |
| Jun 19, 2023 at 22:02 | comment | added | ggorlen | @Braiam When did I say I score 0 as wrong? You're putting words in my mouth--I'm not claiming LLMs are never correct; I'm talking about an overall trend: -10 points, zero positive comments or upvotes, disrespectful comments. Clearly not a positive contribution, but this is literally the first user I found scrolling through my history of ChatGPT flags and it's indicative of what I've seen of ChatGPT usage on SO. I could have picked a "worse" example. Yes, LLMs are sometimes correct, and I'm sure a couple of these are (the Playwright one was--I'm the 8th highest rated user in that tag) | |
| Jun 19, 2023 at 20:48 | comment | added | Braiam | The problem with this answer is mainly that you do not know (and here I'm quoting you "I'm not a subject matter expert on all these topics") but you take 0 scored answers as "being wrong". That kinda defeats the purpose of my question. You are just showing random examples of answers that use chatgpt, rather than from an objective point of view was this answer right or wrong for the question asked. I included an example on my question of chatgpt being right. | |
| Jun 19, 2023 at 16:35 | comment | added | justhalf | @TomWenseleers "Definitely after a couple of prompts, asking some clarification on this or that, testing the code & quickly debugging it if need be, etc." this is definitely a good process leading to a good answer, you're doing well with that. What ggorlen is objecting here is that it's much easier to not do what you did with ChatGPT output and instead just copy paste it, and that's what people apparently (and unsurprisingly) did. | |
| Jun 18, 2023 at 19:32 | comment | added | ggorlen | Nobody's telling you or preventing you from doing that. Go ahead and use LLMs, just please don't feed it back into the human training data. Most people are doing a much worse job of it than you--they're overwhelmingly rep farming spammers rather than well-intentioned university professors, as far as I've seen. | |
| Jun 18, 2023 at 19:25 | comment | added | Tom Wenseleers | Still, something is not adding up - I guess many are just using ChatGPT3.5, whereas I am working with ChatGPT4 and probably asking questions in a better way. But for me the quality of what I get out of it is honestly on a par (if not better) than the average answer I got on SO now (at least in my field)... Definitely after a couple of prompts, asking some clarification on this or that, testing the code & quickly debugging it if need be, etc. | |
| Jun 18, 2023 at 19:21 | comment | added | ggorlen | Believe me, I'd tell you if I thought there was a good GPT answer I've seen. I've seen none that I can tell. If these were even remotely apparent, there'd be no discussion. Given how obvious the copy-pasted ones are, I find it hard to believe that there'd be an equal quantity of GPT answers masquerading as humans that are so good, I can't identify them. In my SME tags, I know almost all of the "regulars" in the tag, and I've seen the quality of their work before LLMs, so most of the good answers are either posted pre-LLM, or on par with the work of trusted users established before LLMs. | |
| Jun 18, 2023 at 19:18 | comment | added | Tom Wenseleers | So with that last comment it makes it clear that your sample is entirely focused on the ones clearly engaging in abuse, whilst you have no idea what % of the good answers have been written whilst being assisted by ChatGPT. If abuse is obvious so much the better: that would also make it easy to take action in an automated way then by the SO system itself... | |
| Jun 18, 2023 at 19:16 | comment | added | ggorlen | If it's undetectable, I don't mind as much. LLMs are useful tools. I use them. I'm not really a purist. But gosh are they ever obvious. Most answers I reference day-to-day are older than 11/30/22 anyway, and I'm considering adding the userscript to block any newer answers. | |
| Jun 18, 2023 at 19:14 | comment | added | ggorlen | It's not cherry picked in the sense that the quantity of posts by users that obviously abuse the system outnumbers the useful GPT posts by a factor of hundreds, from what I've observed spending hours per day on SO since 11/30/22. Your posts are literally the first remotely useful GPT answers I've seen on the site out of hundreds. If that isn't cherry picking, I don't know what is. Like I said, I'll compile all of my GPT flags and post them when I have the chance, so there will be no accusations of cherry picking. | |
| Jun 18, 2023 at 19:11 | comment | added | Tom Wenseleers | It is cherry picked in the sense that it's focusing on users that obviously abuse the system. You won't see all the good answers that are posted, which were inspired or partially based on ChatGPT4 output, but don't disclose it - they are practically undetectable with good reliability anyway... These sorts of accuracy benchmarks should be done blind. That would be a job for SO itself. As for moderator overload: clearly many technical possibilities to reduce that load if they would be open to change, greater automated actions to counter abuse, etc... | |
| Jun 18, 2023 at 19:04 | comment | added | ggorlen | As I said, OP shouldn't have unilateral say over what winds up in a thread. They don't even have say about deleting their own question once it's answered with an upvote, because it's now a community resource (think Wikipedia). Ostensibly, LLMs were allowed by the company because they're worried about abusive deletions, and letting OP be the arbiter of this is a sure way to achieve the exact opposite. Also, as I said in the post, I didn't cherry pick this. This is just the first user I came across in my flag history among dozens. | |
| Jun 18, 2023 at 19:03 | comment | added | Tom Wenseleers | That to my mind is mainly an indication that too much of the responsibility of deleting bad answers lands on the shoulders of moderators. Just let the OP do that, who has a vested interested & clear incentive to see his/her question answered & the problem solved with a good, working solution. For my own questions, I've only seen one bad nonworking GPT3 answer being posted by someone, but I would have deleted that right away if I could - that would have been no work for me... | |
| Jun 18, 2023 at 18:57 | comment | added | ggorlen | Again, the presence of a few good LLM-based answers doesn't justify the unmoderatable barrage of nonsensical crap I've witnessed since LLMs landed. Stand by: I'll be compiling a list of all the posts I've flagged, almost all of which are wrong, unmitigated garbage copy-pasted by someone who probably has no idea what the question is even asking. I don't mind losing the few false positives. We can get those from LLMs directly! I get that you like LLMs and have written good answers with them, but the rest I've seen are rep farming spammers like the featured user in this post. | |
| Jun 18, 2023 at 18:55 | comment | added | Tom Wenseleers | Going forward SO would not "melt down" as a result of poor quality GPT assisted answers. Certainly not if a good reward system is in place and if the OP has better means to up & downvote good & bad answers or delete the crappy ones... It's that human vetting that's currently not functioning very well. That should be based on the quality of the answer alone... And should probably give more weight to actual domain experts etc... That should be fixed... | |
| Jun 18, 2023 at 18:51 | comment | added | Tom Wenseleers | If you call answers like stackoverflow.com/questions/48119360/… and stats.stackexchange.com/questions/76925/… as "polluting SO" then that's quite rich. It's providing in this case an answer that's verified correct by a domain expert like myself, merely assisted by ChatGPT4, to questions where there was previously none... | |
| Jun 18, 2023 at 18:48 | comment | added | ggorlen | Even if you worship at the altar of LLMs, you probably still shouldn't pollute your training data recursively: "What will happen to GPT-{n} once LLMs contribute much of the language found online? We find that use of model-generated content in training causes irreversible defects in the resulting models, where tails of the original content distribution disappear." | |
| Jun 18, 2023 at 18:45 | comment | added | ggorlen | Giving OP the ability to instantly delete any answer is ripe for abuse. That's way too much control over the community resource. It's not a matter of not liking GPT. I use GPT. It's a matter of keeping it where it belongs so it doesn't undermine/overwhelm other avenues to knowledge. As I've said before, if you want a human answer, there should be a place for that. Instead, people like yourself seem to be interested in polluting any and all sources of information with GPT to the point where there's no avenue to access guaranteed human information any longer. | |
| Jun 18, 2023 at 17:13 | comment | added | Tom Wenseleers | I always wanted to delete the nonworking RGL solution suggested to my question as that's a useless comment & nonworking (that was a human-written answer, but same principle applies: it's useless). The working ChatGPT4 based solution that I posted eventually was deleted 3 times. Now I rewrote it a bit to try to keep it up. Still gets downvoted by users or moderators just because they don't like GPT. There is something seriously wrong with the voting system... stackoverflow.com/questions/48119360/… there w | |
| Jun 18, 2023 at 17:09 | comment | added | Tom Wenseleers | In most of these cases the OP correctly detected that the answer was not working, so if they are looking for an easy way to get rid of such spam that would be easy enough: just grant the OP the privilege to delete any nonworking solutions & downvote without it costing reputation. You don't need moderators for that. The OP will be the one with the highest incentive & the best knowledge of whether the answer works or not & solved his/her problem or not... | |
| Jun 18, 2023 at 16:05 | history | edited | ggorlen | CC BY-SA 4.0 |
typo
|
| Jun 18, 2023 at 15:34 | history | edited | ggorlen | CC BY-SA 4.0 |
added 243 characters in body
|
| Jun 18, 2023 at 15:29 | history | answered | ggorlen | CC BY-SA 4.0 |