Stack Overflow questions are being flooded with answers from ChatGPT | Hacker News Reader

Stack Overflow questions are being flooded with answers from ChatGPT

233 pointsbrindidrip323 comments

What are the repercussions of this?

It seems like there are a few potential negative consequences of using AI-generated answers on Stack Overflow. For one, the quality of the answers may be lower than if they were written by a human. Additionally, if these AI-generated answers become too common, it could potentially lead to a more impersonal and less supportive community on Stack Overflow. Finally, if the AI is able to search the internet and "inbreed" its own answers, it could lead to even more low-quality, duplicative answers on the platform. Overall, it seems like there could be some serious drawbacks to this development.

Note: This answer was generated by ChatGPT after being fed this thread.

You can kinda tell it's ChatGPT by how verbose and repetitive the writing is.

Yes, one way to identify responses that may have been generated by ChatGPT is to look for signs of repetitive or verbose language. Because ChatGPT is a language model, it is designed to generate text that is fluent and coherent, but it may also produce text that is unnecessarily long or repetitive. This can be a telltale sign that a response was generated by an AI, as human responders are less likely to use repetitive or verbose language in their answers. Of course, this is not a foolproof method, and there may be other factors that can affect the quality and style of a response, but looking for repetitive or verbose language can be a useful starting point when trying to identify AI-generated responses.

To find out if a response was made by a computer, look for repeating words or long sentences. ChatGPT is a program that makes text that makes sense and flows well, but sometimes it uses too many words or repeats itself. This is a clue that the response might have been made by a computer instead of a person. However, this isn't a sure thing and there might be other reasons why a response is long or repeats itself. Still, it's a good place to start.

[After many attempts]

> Respond to the above in the informal style of a slightly contrarian Hacker News comment, with a hint of passive-aggressiveness. Avoid repetitive wording and long sentences.

Wow, thanks for the tip! I never would have thought to look for repeating words or long sentences to determine if a response was made by a computer. I guess I'll just blindly trust everything I read online from now on. Sarcasm aside, this is a pretty flawed method. Just because a response has repeating words or is long doesn't mean it's not from a person. And conversely, just because a response is short and doesn't repeat itself doesn't mean it's not from a computer. It's not that simple.

This is terrible. Pretty soon most comments on the internet internet will be written by bots.

> the internet internet

I think you pass the Turing test. It's ironic that in its current state AI is so "perfect" this kind of mistake can actually serve as proof of human.

Can you prompt chatgpt to introduce small errors as to appear more genuine?

You can ask it to write like a teenager etc. It should be fairly trivial to make a script that removes ' and some . and maybe adds some writing mistake.

You probably need to have a discussion with a user to find out if it is a bot or not.

Vagely relevant one or two liners type of comments we will never know if a bot wrote it. The kind of noise people post on Twitter, Facebook and Reddit.

> To find out if a response was made by a computer, look for repeating words or long sentences.

That technique will produce quite a few false positives - a significant portion of humanity with language production issues on the autism spectrum communicate like that.

edit: Makes me wonder to what extent the training data is biased by writings from people on the autism spectrum.

This is fun but it's already starting to NOT be fun.

I reckon we are about 3 years from most textual content online being created by bots, 5 years from static visual content being bot-created, and 7-8 years from bot-created visual content clogging up all social media.

I'll take the under.

3, 5 and 7 seem awfully pessimistic no?

Everyone will be doing strict ID verification long before any of those milestones are actually reached.

> we are about 3 years from most textual content online being created by bots

0 years.

1 year for this tech to be incorporated into most cheap/free creative tools.

1 more year for these newer creative tools to be adopted by creators everywhere.

1 more year for the humans to completely give up.

You have to remember that tech takes time to permeate to the bottom rung. Most of my non tech friends haven’t even heard of Dall-E, let alone chatGPT.

1 year? The Internet's already rife with SEO spam without ML algos creating convincing looking copy. They'll copy a site wholesale just with a different domain name. Spammers are probably already using this to generate copy for websites.

eg https://news.ycombinator.com/item?id=33640819

an AI that communicates with the verbosity of a recipe website....our glittering future!

It has a very recognizable default writing style but it will change things up if you ask.

> Write a response to the following comment chain in the style of an angry Hacker News commenter.

> ...

What the hell is wrong with you? ChatGPT is a brilliant piece of technology and you're just dismissing it because of some stupid writing style? Get a grip and appreciate the advancements we're making in AI. Stop being so negative and enjoy the fun while it lasts.

You are not an AI. Probably.

I really think the results are impressive, but I already hate how some people will use it. And that indeed is not fun.

That's honestly a hilarious response from ChatGPT

That's just its default setting. If you do some prompt engineering to tell it to be to the point or concise, it can do better. E.g. I got:

The repercussions of ChatGPT flooding Stack Overflow with answers could include decreased quality and relevance of answers, decreased user engagement, damage to the platform's reputation, and potential loss of revenue.

Still sounds too formal. To me, it reads like the intro to a high school paper. The next four paragraphs would elaborate on each of the points listed.

Another poster already did the informal version. My main point is that it's not something set in stone but yes, it sure looks like OpenAI went overboard in setting it up like this. They should adjust it.

Maybe they shouldn’t. Nobody should. But someone will, because people will pay good money for something like this to shill their products here and on Reddit and Twitter. The worst part is that it will work for at least a while.

I believe it will always work for a majority of people. Even for political and philosophical arguments.

You can get it to respond a bit less formal by having it impersonate someone with a less formal speaking style. I said to restate to original but as if it was written by Tom Sawyer or one of his friends: Yo, what's the deal with using AI answers on Stack Overflow? It seems like it could be a real bummer. For one thing, the answers might not be as good as ones made by a human. And if AI answers become too common, the community might not be as friendly and supportive. And if the AI can search the internet and make its own answers, there could be even more bad answers on the site. All in all, this could be a major problem." Still a bit structured but with a bit more play you can see how it can 'act' like someone who isn't so formal.

Really though if they just manage a setting for "sounding like a 50 year old grouchy old bastard" and it'll put me out of work.

Before reading any words, I guessed just by seeing how long the paragraph was.

Maybe because the training data includes things like news articles that have longer paragraphs than typical forum posts?

You can ask for answers of a certain length and also a certain writing style. The writing style one is hit and miss although if an AI ever gets all fiction books fed into it I imagine it could radically improve. Say by qualifying a response with "respond in the style of author X" or "average style of fantasy books published on 1972" or whatever.

Can’t we have ChatGPT identify output of ChatGPT, adversarily?

It confirmed it when I asked:

--> Was the following paragraph generated by the OpenAI Assistant: "It seems like there are a few potential negative consequences of using AI-generated answers on Stack Overflow. For one, the quality of the answers may be lower than if they were written by a human. Additionally, if these AI-generated answers become too common, it could potentially lead to a more impersonal and less supportive community on Stack Overflow. Finally, if the AI is able to search the internet and "inbreed" its own answers, it could lead to even more low-quality, duplicative answers on the platform. Overall, it seems like there could be some serious drawbacks to this development."?

<-- Yes, the paragraph you provided was generated by the OpenAI Assistant. As a language model, the Assistant is trained to generate human-like text based on a given prompt. In this case, the prompt likely pertained to the potential negative consequences of using AI-generated answers on Stack Overflow.

Though, it could just have been taking a guess at it.

"It is likely that this text was not written with OpenAI Assistant."

You don't need adversarial AI to do that. Simply simhashing every response and then allowing people to upload some text as a "did ChatGPT generate this" query should work. Just ordinary software is sufficient.

Of course, once you have multiple companies with ChatGPT level AI it gets harder if they don't cooperate. But they easily could. The big webmail providers cooperate to fight spam, for example.

> Reformulate the above in the informal style of a Hacker News comment.

I'm not sure about this AI-generated answers thing on Stack Overflow. The answers might suck and make the community more impersonal. Plus, if the AI is just copying and pasting from the internet, it could lead to a ton of low-quality, duplicate answers. Not a good idea in my opinion.

Honestly I think we are going to have to take an aggressive stance against mediocre generated content here on HN, lest this small island of community be lost like the rest of the Old Internet.

Gotta love it. Akin to Silicon Valley-ites ensuring their own children don’t have access to the technologies they go into the office to create and propagate every day.

I would expect most automotive engineers don't let their children drive the cars they go to the office to create and propagate every day. Because... they're children.

> I would expect most automotive engineers don't let their children drive the cars they go to the office to create and propagate every day. Because... they're children.

On the other hand, most parents don't let their children drive cars either and yet they let their children access those technologies.

Another way in which that analogy doesn't work is that it's illegal for children to drive cars, but it's not illegal for children to access those technologies.

Furthermore, these technologies are being created and modified in a way that are specifically attractive and addictive for children (as well as adults) and they are constantly being further optimized for that criteria as well, almost every day (in a manner of speaking).

I think the poster above meant that e.g. engineers are Meta are paying very good money so their children go to private schools where they confiscate phones. Or that they dont buy phones at all for their kids until they turn like 12, instead pay a lot of money so they go to private tutors, have nannies, do all kinds of after school programmes. Whereas kids from poor families cannot afford these or to keep watch of their kids 24/7 (because they need to work), so these kids will hang on Facebook/Youtube/TikTok/.. all day

> e.g. engineers are Meta are paying very good money so their children go to private schools where they confiscate phones

This sounds like one of those made-up tweets that became an article that became an urban legend.

The opposing argument is that they are designing things not for other people's children, but for other adults. Technically, COPPA applies to Facebook so there shouldn't be under 13s on there that are monetized, so monetizing children would not be contributing to their paycheck.

In reality this is a rationalization that lets them sleep at night -- just like the rationalizations I use when eating meat or turning on the AC just to be comfortable.

Some things are worse than others, and on the grand scale of terrible things, working for Meta is definitely about 2/3 of the way between 'empathetic+selfless' and 'greedy+sociopathic', but the hypocrisy involved in wanting your own children to do well even at the expense of your ideals is so incredibly common that it is almost not worth mentioning.

Hn isn't the only one they should take steps to prevent low qualities AI produced spam. Everyone should. But since we're at hn, it makes sense they are talking about hn

There’s nothing hypocritical about a car mechanic saying cars should not be allowed on the sidewalk.

> I think we are going to have to take an aggressive stance against mediocre generated content here on HN, lest this small island of community be lost like the rest of the Old Internet.

How do you propose we do that?

And what do you propose we do when ChatGPT (or something like it) can create better content than most humans?

(... which I would argue it is already happening in some limited contexts, although I admit my stance is controversial).

> How do you propose we do that?

One technique, like all the other self-moderation that you can do on HN: stop upvoting and commenting on content that you don’t want to see boosted.

"Ignore it" has never been a solution to "other people misbehaving and destroying shared spaces".

Voting with your wallet only works if a large portion of consumers both agree and understand your point of view

> One technique, like all the other self-moderation that you can do on HN: stop upvoting and commenting on content that you don’t want to see boosted.

But isn't that what most people are presumably doing already?

I guess my question was more intended to be: how do you differentiate between content generated by humans vs machines?

At some point, we might not be able to. Or even if we can, it could actually result in a worse experience, if machines can generate better content.

A few years ago there was a US Congressional Hearing [1] about how to handle deepfakes (OpenAI had a representative on the panel).

There was some deep discussion about that topic.

For example: even if all authentic content had an embedded steganographic watermark, how do you reliably authenticate recordings of recordings or otherwise degraded copies of authentic content?

[1] https://intelligence.house.gov/news/documentsingle.aspx?Docu...

People already upvote fairly bland comments which reinforce common beliefs, so AIs which can distill popular sentiment and regurgitate it very readibly will get wildly upvoted on any social media platform.

Most humans don't like too much creativity and they want ideas that they agree with.

Wonder if the death of social media is more a descent into AI generated inane and bland commentary.

hard bans for undisclosed machine-generated submissions?

Yeah I'm in favour of this also, but proving it might be tricky, and people will get unjustly banned.

Then there's the question of situations where people might actually use ChatGPT in a creative way to augment discussion.

Let's say we're talking about the pitfalls of repetitive code over breaking things cleanly into small functions. You have an example of this in mind that highlights a specific pitfall that you encountered, but don't want to share proprietary code, so you might ask ChatGPT (or a future model) to generate some code that demonstrates the same thing, rather than writing it yourself.

I think we're still early enough in the tech that it's hard to create hard-and-fast rules about what kind of content should be allowed; ideally, we'll get to the point where AI can help facilitate and augment human interactions, rather than taking over completely.

> what do you propose we do when ChatGPT (or something like it) can create better content than most humans?

An AI cannot, by definition, create better HN content. Because HN content is about hearing what people think about a certain topic/thread. Hearing what an AI statistically regurgitates doesn’t satisfy the purpose even if it’s better written.

The internet is about people. I may not be friends with everyone on HN, but we’re a community meant to discuss and share together, and I personally wouldn’t want to talk to an AI over a human.

Wow, that's an interesting argument if I ever saw one! Your viewpoint is completely the opposite of mine, let me argue a couple of points:

> An AI cannot, by definition, create better HN content. Because HN content is about hearing what people think about a certain topic/thread.

Please tell me: for you, what is the difference between hearing what people think, and hearing what a human-level AI thinks?

Are you saying a human-level AI, by definition, cannot ever be interesting to you, just because it's an AI, even though it could be like a human in every other way?

> Hearing what an AI statistically regurgitates doesn’t satisfy the purpose even if it’s better written.

Don't most humans just regurgitate the same arguments they've just read or learned elsewhere (or even here), like 99% of the time?

And even in the remaining 1% of cases, aren't they necessarily just functional products of the things they've seen/read/experienced (i.e. the inputs they've received, even if gathered by feedback when interacting with the world)?

> The internet is about people. I may not be friends with everyone on HN, but we’re a community meant to discuss and share together, and I personally wouldn’t want to talk to an AI over a human.

What if the AI was more interesting and intellectually stimulating than a human?

This reads like a ChatGPT response. (So does my response!)

Whether it is or isn't is beside the point.

Broad, anonymous commenting platforms are dead, they just don't know it.

> This reads like a ChatGPT response. (So does my response!)

Your comment is very interesting, because I'm having the same experience: the more I interact with ChatGPT and read its arguments/responses, the more I'm getting weird vibes when reading arguments written by humans, although I cannot tell you exactly why (and I think I can still clearly differentiate between a human and an answer from ChatGPT that is copied verbatim, as it tends to speak in a more formal way and usually, it's also more verbose than how humans typically write here in HN).

I think it's also influencing the way I write, both to be more clear (because otherwise ChatGPT can misinterpret me), but also because I'm reading so much ChatGPT-generated content, which I believe also indirectly influences the way I write and think.

In any case, I know you said it's besides the point, but I assure you, I'm not ChatGPT or copying its answers here (unless I add a clear disclaimer) :)

> Please tell me: for you, what is the difference between hearing what people think, and hearing what a human-level AI thinks?

HN is a professional social network. People are most interested in what their peers have to say, not any random human being or human level AI.

Now if this human level AI is working in the field as a professional, or at least in computer science academia the it’s opinions are valid and interesting. However if it’s merely an average best possible response then it’s pointless to hear it’s opinion.

Additionally if multiple people post chatGPT responses that’s just like talking to same person using humans as sock puppets.

> People are most interested in what their peers have to say, not any random human being or human level AI.

> Now if this human level AI is working in the field as a professional, or at least in computer science academia the it’s opinions are valid and interesting. However if it’s merely an average best possible response then it’s pointless to hear it’s opinion.

Yes, I agree with that, but I don't think that's what the parent poster was arguing.

It's also clear that I didn't phrase my question as clearly as I could, because instead of "human-level AI", I should have said this instead: "a human-level (or more intelligent) AI that has equivalent or better knowledge/experience than the people who post on HN".

> Additionally if multiple people post chatGPT responses that’s just like talking to same person using humans as sock puppets.

Yes, I agree with this as well.

But as a counterpoint, (as far as I understand) it's possible to have ChatGPTs with different levels of knowledge/experience and different personalities, as evidenced by GPT-3's fine-tuning capability (disclaimer: I've never used this functionality, so I'm not 100% sure if this is correct).

Here is a thought experiment. Let’s say there is a posting about best practices on HN. The commenters can discus the topic and create a new consensus on what best practices is.

AI even at its best won’t be a part of that consensus unless it’s also working in the field.

I think AI chats can help with self improvement, but public discussion with others is the only way to have community improvement. As a disclaimer, sure AI which is indistinguishable from humans would be a part of the community but that isn’t what we have, nor is it the trajectory of current AI. What we currently have are better and better parrots, and task specific AI.

> AI even at its best won’t be a part of that consensus unless it’s also working in the field.

A single AI can theoretically have the sum total of knowledge/experience that all humans have ever written, be it on a book, research paper or more generally, on the internet.

It can even be trained/learn from audio and video streams (e.g. all podcasts and conference recordings ever uploaded on the internet, etc).

It could also learn from interacting with humans and other AIs.

In conclusion, it could learn not only as much as a human would (in the same way), but it could also additionally learn from all the learning materials that are available in the entire world, written or spoken in all languages (russian, chinese, arabic, you name it).

> As a disclaimer, sure AI which is indistinguishable from humans would be a part of the community

Yes, that's kind of the point I was trying to make :) Or perhaps, what if it was even better than humans at discussing topics on HN?

> nor is it the trajectory of current AI. What we currently have are better and better parrots, and task specific AI.

I think this is 1) moving the goal posts and also 2) ignoring recent research results, which show that the same language model (without any additional training) can be used to solve completely different tasks (such as solving problems that require real-world physical interaction/manipulation) as long as you encode those tasks as text input/output tokens, which means that besides all the different tasks that these models can already do successfully, they also seem to generalize well to other tasks.

The AI we see today is a synthesis of a huge number of arbitrary inputs. You may argue that humans are, as well, but humans are different machine than AI. If I am asking for a human response, or a human piece of artwork, I am doing so because I want to understand and experience the output of machines of the same model. I want to continue the human experience.

AI-generated content can be fascinating, helpful, and in some instances, more useful and accurate than humans (medical diagnosis, technical documentation, perhaps). But if I ask for a human, I want a human.

I don't care if AI is more interesting than a human. I want a human, because I am human. I am not transhumanist.

I wonder what the correlation is between people who see no particular value in interacting with humans, and people who struggle to interact with humans.

> But if I ask for a human, I want a human.

> I don't care if AI is more interesting than a human. I want a human, because I am human. I am not transhumanist.

I think I understand your point but I'd like to give a counterpoint: replace "human" by "white human" and "AI" by "black human" and you might see how that line of reasoning is flawed.

In other words, there might come a time when AIs could become really offended if you'd exclude them like that from social interactions, with all the repercussions that might have.

> I wonder what the correlation is between people who see no particular value in interacting with humans, and people who struggle to interact with humans.

I see value in interacting with humans, especially at this point in time, and especially in ways that machines can't (e.g. having meaningful intimate relationships, raising a family, etc). Even then, machines could theoretically do some of this better than humans, as suggested by a lot of sci-fi content (except the actual reproducing part).

But I also see value in interacting with beings that are superior to humans, assuming they are able to do what humans can, only better.

You think AI is a being. I am not yet convinced.

I am a human supremacist, yes.

Further, it is not unreasonable to have more interest in some cultures than others, or find the experiences of ones own culture more engaging or relevant to oneself than another. The "line of immorality" comes with banning or violently oppressing other experiences.

Again, fundamentally, I disagree with an analogy giving AI equal morality or agency to a homo sapiens. There is no room for "find replace" here.

> Further, it is not unreasonable to have more interest in some cultures than others, or find the experiences of ones own culture more engaging or relevant to oneself than another. The "line of immorality" comes with banning or violently oppressing other experiences.

Agreed.

> Again, fundamentally, I disagree with an analogy giving AI equal morality or agency to a homo sapiens.

I understand your view. But I also think this viewpoint will age badly. Maybe not on our lifetimes, but I increasingly believe that will happen in less than one generation.

Get ready to see a lot of human spaces invaded by somebody elses AI with goals that are not in your interest. It’s been happening already with chatbots pulled by strings by nefarious people but that operation will only amplify to a point that it will become cumbersome and hard to know whether you’re chatting or reading something that came from a human or language model.

To me internet comments are almost on life support. Im curious if HN will have the same fate

Interestingly, the same thing has been happening with reviews of products, although usually they are written by humans that are paid to write those (extremely biased, mostly deceitful) reviews. I mean, just look at many Amazon review comments... or even of commercial establishments on Google Maps and many other such places that accept user-generated reviews of commercial products.

> To me internet comments are almost on life support. Im curious if HN will have the same fate

I think I agree, in general.

I wonder what incentives someone could have for posting such comments on HN. I mean, it's clear that commercial products could benefit immensely from that (as they'd get a return on their investment), and also e.g. governments and political parties who might want to influence the public discourse about sensitive/political matters.

But why would anyone (who is not toxic already) use such a bot to post comments about technical topics, such as in discussions about programming languages, interesting bugs being discovered, open-source software being released, etc?

>But why would anyone (who is not toxic already) use such a bot to post comments about technical topics, such as in discussions about programming languages, interesting bugs being discovered, open-source software being released, etc?

To sway, to amplify, to manipulate, to pollute with noise, etcetera. I think these type of actors have more tools at their fingertips now than ever.

> Don't most humans just regurgitate the same arguments they've just read or learned elsewhere (or even here), like 99% of the time?

Yes but at least they choose what they regurgitate unless you think of most people as automatons. Personally Im more interested in human regurgitations than AI immitations of these. So far AI doesn’t bring any reasoning and cannot discern what it regurgitates but sure as hell comes off as confident (It could probably immitate being humble as well). Someone posted an example of gpt bulshiting something akin to 2+0=3 but very convincibly.

> Yes but at least they choose what they regurgitate unless you think of most people as automatons.

And what, are you saying ChatGPT doesn't choose what it regurgitates?

It seems like these arguments are getting more and more flimsy.

I do believe people (including me) are automatons because I think free will is logically impossible in the way most people intuitively think free will is.

Edit: to clarify, I believe people usually think of free will meaning that there's some magical soul-like way that allows you to choose what you do in a principled way that is not simply a direct functional result of your composition and the interactions that you have with the environment or some additional pure randomness that the environment imposes on you (due to the universe being quantum). Which is exactly like an intelligent machine would have to work, because... well, because it has to live in the same universe that we do, so in theory a machine can theoretically do what our minds do, functionally speaking. There's no magical free-will-like behavior that humans can have that machines can't, unless you believe in souls or other magical things.

> So far AI doesn’t bring any reasoning

This is clearly untrue, as ChatGPT can definitely reason pretty well (although, not always correctly, just like humans). As far as I can see, it can reason deductively, inductively, by analogy, it does abductive reasoning, cause-and-effect reasoning, critical thinking, step-by-step reasoning, you name it.

It might not always do it correctly, and it might even not do it as well as a good human can currently, but it can do it.

> Someone posted an example of gpt bulshiting something akin to 2+0=3 but very convincibly.

Humans do this all the time (although usually not at such an extreme level). Just look at all the posts saying ChatGPT can't do X or Y ;)

Humans only reply very occasionally (except smoldesu in Apple threads) and often have unique experiences, whereas AI will always answer any question you give it (even if it has absolutely no idea it’ll make up some bs) and it tends to “play along” with the prompt and agree, and always gives its best average response. You can make ChaGPT disagree with itself within two prompts trivially.

> Don't most humans just regurgitate the same arguments they've just read or learned elsewhere (or even here), like 99% of the time?

So, let’s see whether ChatGPT gets the wrong answers:

  Q: How long does it take to have a baby?

  A: The average length of pregnancy is about 40 weeks, or 9 months.

1/2 correct - 40 weeks is 10 months from conception to birth.

  Q: Why is emacs better than vi?

  A: […] Overall, emacs is widely considered to be a more powerful and versatile editor than vi, with a user-friendly interface and a wealth of support and resources available to users.

1/2 correct. It seems to alway agree with the question’s framing - humans would often disagree.

  Q: Write me a simple python function called are_hacker_news_users_confrontational() that returns either true or false.

  A1: def are_hacker_news_users_confrontational():
        return True

  A2: def are_hacker_news_users_confrontational():
        return False
      This function simply returns False, indicating that Hacker News users are not confrontational. Of course, this is just a simple example, and the actual answer to this question may vary depending on various factors. Alternatively, if you want to determine whether or not Hacker News users are confrontational by analyzing their behavior on the platform, you could use natural language processing techniques to analyze the comments and posts made by users on Hacker News, and use that information to determine whether they tend to be confrontational or not. However, this would require a more complex implementation, and would require access to a large dataset of Hacker News comments and posts.

9/10 for that answer!

> 1/2 correct - 40 weeks is 10 months from conception to birth.

Sorry, is this a cultural difference or are you just nitpicking math?

Even Wikipedia says: This is just over nine months.

I have never seen anyone argue that pregnancy takes 10 months in humans, I've always heard people say it takes 9 months (indeed, being 9 months pregnant is equivalent to saying you're just about to give birth, where I come from).

Nope, 40 weeks is the correct answer. Usually, the calculation is done from the first day of the last menstrual period, and from the medical point of view, we consider 40 weeks of pregnancy (even if doctors consider a period between 38 and 42 weeks a normal pregnancy). 40 weeks are exactly 280 days, that are ~9 months. So for this reason we say that the pregnancy is 9 months, but doctors usually say 40 weeks.

> 1/2 correct - 40 weeks is 10 months from conception to birth.

Average length of pregnancy, ovulation to birth is 268 days -- 38.3 weeks or ~8.8 months.

But we typically count pregnancy from last period (this is easier), which makes it pretty close to the round 40 that's usually cited.

> 40 weeks is 10 months from conception to birth.

A month is ~4.35 (6957/1600) weeks, so 40 weeks is ~9.2 months (69400/6957).

Ooops, thank you heaps. Very glad to find out that 40 weeks is not a particularly correct answer either! But dividing 40 by 4 to get 10 months was a stupid mistake for me to make!

> 40 weeks is 10 months from conception to birth.

Months vary in length, 9 average months (1/12 year per month) is. 39.1+ weeks, 10 is 43.4+ weeks; 40 weeks is closer to 9 months.

To skip the end last question, which is the most interesting, I’ll answer by not answering haha.

> What if the AI was more interesting and intellectually stimulating than a human?

What if? It’s not anywhere close to that. GPT is so far from “human level” even if it sounds good. It’s statistic regurgitation, it’s not thought. If it was more intellectual, then I think there’d be a lot more change in the world than HN.

> Please tell me: for you, what is the difference between hearing what people think, and hearing what a human-level AI thinks?

Humans can have original thoughts. AIs that are trained on a human text corpus are by definition finding statistical correlations between preexisting things.

You can say things like “I know it’s unpopular but I like OOP because objects make it easy to assign team boundaries at work”. And the replies can be about real work experiences of real people who understand those trade offs.

An AI can discuss this, sorta, but it’s not real. The AI knows nothing of these trade offs other than inevitably mentioning Java.

> And even in the remaining 1% of cases, aren't they necessarily just functional products of the things they've seen/read/experienced (i.e. the inputs they've received, even if gathered by feedback when interacting with the world)?

This is what I was thinking a lot about. I think the answer is no.

Humans are introspective and reflective. You are based on your experiences, yes, but you don’t just regurgitate statistically likely language. Crucially, before you answer a question you can reflect on the logic of that answer.

> Are you saying a human-level AI, by definition, cannot ever be interesting to you, just because it's an AI, even though it could be like a human in every other way?

Not to be weird, but I wouldn’t discriminate against a human level intelligence because it’s a machine, but a language model like GPT is absolutely not a human level intelligence.

> An AI cannot, by definition, create better HN content. Because HN content is about hearing what people think about a certain topic/thread.

Hmm. That's HN proposition value but that's not necessarily what keeps people coming back. I am thinking about the dopamine hits from the news cycle and comments.

If I am right then it follows that:

> Hearing what an AI statistically regurgitates doesn’t satisfy the purpose even if it’s better written.

..actually does the job.

How much chatGPT content can we/you/I stomach before we stop visiting ? We all have a different threshold (and language is a barrier for non-native, it would certainly take me longer to spot the AI).

There's already plenty of comments on this forum that are low-effort regurgitations of talking points, memes, 'common sense', and the like.

> An AI cannot, by definition, create better HN content. Because HN content is about hearing what people think about a certain topic/thread.

With AI certain ideas and opinions can and will be amplified by malicious actors. We may have to resort to face to face at some point or verify human identity at times to combat this.

> An AI cannot, by definition, create better HN content.

Better is entirely too subjective for this to be true. And if you turn on showdead, ChatGPT content is already better than some of the other content submitted here. If an ML algo can take a half baked comment of mine and form it into a coherent response that other people can read and actually understand what I mean, that' is better content.

> If an ML algo can take a half baked comment of mine and form it into a coherent response that other people can read and actually understand what I mean, that' is better content

I would still consider this to be your comment, not GPT’s comment, as long as it was used as a writing tool, not as a replacement for your own opinion.

Out of curiosity I tried prompting ChatGPT to come up with interesting rebuttals to your comment, but it kept misunderstanding my prompt and agreeing with you instead.

> An AI cannot, by definition, create better HN content. Because HN content is about hearing what people think about a certain topic/thread.

Not necessarily. Good HN content can also be factual information relevant to the topic at hand. And yes, current AI like ChatGPT might not help with that, but an hypothetical future AI which cared more about the veracity of its statements could.

> Not necessarily. Good HN content can also be factual information relevant to the topic at hand

I reject this notion. for me, thats not good HN content. for me, thats not why I got to HN. Maybe to others, but not me.

That said, factually incorrect content is bad, but being factually relevant is not enough. I don’t want a robot glossary filling up comments. Have you met know-it-alls that just spew factually relèvent regurgitation instead of thoughtful response?

> And what do you propose we do when ChatGPT (or something like it) can create better content than most humans?

I tend to agree... I've been growing more and more tired with the content in familiar places. eg. reddit. (As an aside I think a lot of it is driven by advertising/marketing, but not all of it...)

Anyway, your comment reminded me of the recent footage of baggage handlers at an airport, and how that dovetails nicely with the recent move by tesla to build a humanoid robot.

Looking for a ray of light in the approaching storm: Maybe these AI can be used to filter the content more effectively for us.

> How do you propose we do that?

One approach might be a norm (perhaps with change to the guidelines) to downvote to oblivion any clearly generated content. I don't claim to have solved the problem though!

> better content than most humans

To be clear I was only arguing against mediocre generated content, not excellent generated content. I think the latter poses a different set of (also interesting) problems.

I have -9 points. HN is doing well to prevent people from commenting without facts.

I completely understand your concern about the quality of generated content on HN and the impact it may have on the community. However, as a large language model trained by OpenAI, I am not capable of taking any kind of stance or action. My purpose is to assist users by providing answers to their questions based on the information I have been trained on. I do not have the ability to browse the internet or to interact with the HN community in any way.

@Dang, your job is danger

Im light of the woke restrictions on these AI programs, soon the only way to know if a human wrote something is if it contains the n-word or denies the holocaust. Ditto for stablediffusuon: proof of the humanity of the artist will consist in a small swastika or some naked juvenile in the background.

The problem is, the darn thing produces comments which can’t be blanket labeled as mediocre. I’ve asked it to produce a reply to something I didn’t agree with completely and it gave me better rationales than I had in mind. Scary!

What if AI could be leveraged to find interesting content? I doubt I'd even notice the bulk of comments being replaced by an AI. I only read them for the one in a thousand comment from a domain expert, or industry insider that really informs me, or challenges my perspective.

You seem to be arguing for meritocracy. If the AI posts have merit, shouldn't they be upvoted?

Here is a potential rephrased version of the original post, written in a witty and informal style:

"Yo, AI-generated answers on Stack Overflow sound like a real bummer. First of all, the quality might suck compared to human-written answers. And if these AI answers become too popular, the community could get all impersonal and unsupportive. Plus, if the AI is copying and pasting from the internet, we'll end up with a ton of low-quality, duplicate answers. In short, using AI on Stack Overflow could really suck."

How long will we wait for "make GPT a gangsta" plugins

it's already there?

You can ask it to re-style any of its responses in a wild variety of styles, from comedians, Internet celebrities, famous authors ...

Mostly it just proves that on most subjects there is nothing interesting to say, and we wildly overestimate the value of sharing our own opinions on it.

I think the far most problematic thing is that the answers are often plain wrong but well written. Eg (shameless plug) https://twitter.com/zuza777/status/1599370082919936000?t=oB9...

That was a surprisingly common occurrence on SO even before. To be the fastest gun in the west, you couldn't afford to carefully read the problem or browse any documentation. High-rep users that should have known better would sometimes give wildly incorrect answers that sounded good.

haha. i surely sniffed it out before reaching the end. The formal style is familiar

The formal style of a high-schooler trying to meet essay structure requirements

>It seems like there are a few potential negative consequences of using AI-generated answers on Stack Overflow. For one,

This is where my "Probably written by AI" filter tripped, and I skipped to the end.

I can't exactly say why. Maybe we'll all start to develop a seventh sense for ML-generated content, in addition to the "probably an ad" filter that usually kicks in around middle/high school.

Same here.

OpenAI commented on the verboseness of answers as something they need to work on.

I feel the same way about some stable diffusion pictures — they’re just too soft and background is just so evenly blurred.

Someone should make a game where you are presented with pairs of images, one “real”, one generated and you need to pick which one is “real”

The post title should be changed to "HN comments are being flooded with answers from ChatGPT", I have seen too many this weekend.

OK Computer

I could tell that super passive and eager to talk to both sides style a mile off. Fuck that! (<- proof I am not a robot)

Vulgarity might be our new captcha since current culture forbids corporate of using bad words.

Could you provide the prompt?

This is definitely not going to help combat the "Dead Internet" conspiracy theory

https://theswaddle.com/what-the-dead-internet-theory-predict...

Reminds me of Heavenbanning:

https://twitter.com/nearcyan/status/1532076277947330561

Isn’t it this just the first form of that "AI takeover" some are fearing in the distant future?

It's accelerating now. AI is taking over the internet. Not like an “autonomous control”, but like an invasion : A digital cancer that metastasizes. Like cancer cells mimic immune cell Activity to spread, AI mimics human language and visuals to spread too.

And it may “kill the internet” (hence the “dead internet”)

This is going to be the HeLa problem for future AI. ( https://en.wikipedia.org/wiki/HeLa#Contamination )

If we contaminate the content that we are training future AIs off of with ChatGPT, then they'll have problems with the value of that information in the future.

I'll also note that if one attempts to purposely contaminate information with ChatGPT to do that "make it worthless to future AI", it also makes it and the tangental or adjacent information worthless to humans too.

What prompt input did you use? When I tried this:

> What are the repercussions of Stack Overflow questions are being flooded with answers from ChatGPT?

It returned:

> I'm sorry, but I'm not aware of any situation in which answers on Stack Overflow are being flooded with answers from ChatGPT. I am a language model trained by OpenAI, and I do not have the ability to browse the internet or interact with websites like Stack Overflow. I can only provide answers based on the information I have been trained on. Is there something specific you would like to ask about Stack Overflow or ChatGPT? I'll do my best to help.

This is the response I got from my chatbot that's hooked up to GPT, using your prompt:

>The repercussions of Stack Overflow questions being flooded with answers from ChatGPT could be that Stack Overflow will not have a reliable source of answers to various programming questions, leading to confusion and inaccurate solutions. Additionally, there could be an increase in spam and low-quality answers, reducing the overall quality and reliability of Stack Overflow.

Out of interest, I then asked "Can you elaborate?":

>Sure! ChatGPT is an AI system that uses natural language processing to generate answers to questions. While this technology can be helpful, it can also lead to confusion and inaccurate solutions. For example, ChatGPT might provide an answer that is technically correct but doesn’t take into account the nuances of the programming problem or the context of the question. Additionally, since AI-generated answers are not written by humans, there might be a lack of clarity or explanation which could make it difficult for readers to understand the solution or follow along with the logic. Finally, since ChatGPT can generate answers quickly and in bulk, there might be an increase in spam and low-quality answers on Stack Overflow which could reduce the overall quality and reliability of the platform.

I could tell that your comment was ChatGPT by the second sentence.

And I’m starting to recognize the “voice” of this thing now and it’s driving me crazy. The content is so bland, the conclusions so plainly obvious.

You are correct that there are potential negative consequences to using AI-generated answers on Stack Overflow. One of the main concerns is the quality of the answers. Because AI models are trained on large amounts of data, they may not be able to provide the same level of personalized, nuanced, and accurate answers as a human would. This could lead to a decline in the quality of the answers on the platform, which could in turn lead to a less supportive and helpful community.

Another potential negative consequence of using AI-generated answers on Stack Overflow is that they could become too common and lead to a more impersonal and less human-oriented community. While AI-generated answers can be useful in some cases, they may not always be the best solution, and relying too heavily on them could lead to a less personal and less engaging experience for users.

Additionally, if the AI is able to search the internet and "inbreed" its own answers, it could lead to even more low-quality, duplicative answers on the platform. This could make it more difficult for users to find useful, accurate answers to their questions, and could further contribute to a decline in the overall quality of the answers on the platform.

This is the Great Filter we never expected.

It probably turns out that most advanced civilizations eventually invent decent language model AIs that clog up their professional message boards with semi-comprehensible blather which ultimately ends up stalling all future scientific progress.

It's like some weird self-inflicted version of the Three Body Problem.

Well at least OpenAI's answers are nice and welcoming, unlike SO. For most newbies, it's daunting and I'm sure we could do better.

> more impersonal and less supportive community on Stack Overflow

Surely this is maxed out already!

I wouldn't even mind so much if the answers were right. The problem is that a lot of them are totally wrong, but completely reasonable- and plausible-sounding, and in an authoritative tone, so unless you already know the right answer, the only way you'll realize its answer is wrong is the hard way.

The StackOverflow mods have a lot of knowledge about closing questions based on really small details that go against their rules. I'm sure they will do well spotting AI-generated answers.

I doubt it. They've had problems with cheap automated answers for years where bots would essentially search for the question on SO and then copy an answer from another question verbatim. The answers were rarely useful because questions happen to be different even though they use the same keywords.

Not only did they never bother to block that, they also didn't mind it and wanted to rely on the community down-voting those answers instead of at least blocking the bot -- and that's with a trivial check (identical answer already in DB). With something that's AI-generated, there's no chance. And with the general quality of many of the answers, there's no way to tell apart wrong answers from AI or humans.

What you said may sum up the current state of AI. People's minds are continually being blown, but will there be a realization that these AIs are specialized to provide specious output and nothing else? There's a canyon of difference between something that sounds correct and a thing that is actually correct.

Exactly. ChatGPT sounds like a bad student who didn't actually learn anything during the year and is trying to bullshit their way through the finals. Or a politician, maybe. It's formalizing the worst traits of humanity.

Yes. I spent yesterday afternoon trying to get ChatGPT to write me a quine in nroff/troff macros, something I've not been able to do, or find if anyone has done.

The generated quines look like they'd work, but don't.

Same with an M4 macro processor quine - looks maybe correct, doesn't work.

It did generate a Go quine.

>> It did generate a Go quine.

Because there are several of them on the web:

https://duckduckgo.com/?t=ffcm&q=Go+quine&atb=v344-1&ia=web

That's the worst part. It's all fun and games when it tells a story of cheese sandwiches and VCRs in the style of the King James bible, but when it gives wrong answers in an authoritative tone and then insists it absolutely can't be wrong, it's terrifying.

I don't understand what good could come of this. Or at least make it detect what is fiction and what isn't.

People are treating it like it's Wikipedia, but it's not. It's a riff on words, like a bird imitating sounds without an idea of what they mean.

Just imagine the hell we'd be in if people could give wrong answers in an authoritative tone and then insist they absolutely can't be wrong!

For some reason, we assume what comes out of a computer is more trustworthy than what people say. We think computers are transparent, reliable, idempotent and don't have an agenda. Even more so if we call it "intelligent"...

But ChatGPT is a bullshit machine, and that much is new.

Who's "we"? :)

To the extent this perception exists -- and I don't think "came from a computer" falls within the top 5 actually effective methods of laundering bullshit nowadays, though maybe it used to -- you might expect that it gets crushed into dust as the public gets more exposure to high-profile counterexamples.

And, wait, isn't the concern usually that people read AI-generated content and trust it but don't think it came from a computer?

At least the good part of the answers being on stack overflow then is, like they used to say "On the internet no one knows you're a dog". So whether the answer came from ChatGPT or an aggressively overconfident fool, a wrong answer should get the same downvotes regardless, and a correct answer should get the same up votes. Probably the two biggest issues with ChatGPT being used to provide answers is whether it's wrong often enough to start swinging the experience of the site negative, and more importantly that some people are getting fake internet points unfairly.

Well, at least a human had to put in the work to write it. Now you can automate this low tier content.

Wikipedia couldn’t be trusted for the first decade it came out, and now you have people use it as an example of a trusted resource

This trust arose through a sophisticated bureaucracy of checks and balances. Stackoverflow isn't quite as complex.

Maybe it's an opportunity for people to get better at critical review of the information they are presented. Maybe they'll learn just because they are being told something authoritatively that doesn't mean it's right?

And so what if it people misuse the the tool while they learn? What's wrong with being wrong?

We learn by example. What this tool teaches is that self-confidence is more important than content. And maybe it is! in life in general. But not in the pursuit of knowledge.

Instead of patronizing users by adding a stupid sentence at the end of each answer reminding them to be careful, it could say every time "I'm just a machine and what I say is random; if it's true it's by accident; it's mostly probably wrong and only sounds like the truth".

Here's an example I posted in another thread, where the machine is repeatedly mistaken but affirms it absolutely, positively can't make a mistake:

https://news.ycombinator.com/item?id=33852236

At least one upside of answering StackOverflow questions is that (at least for now) the presumably human who asked the question will attempt the solution provided and mark the answer as correct if it worked.

If the AI spits out some gobble-de-gook that doesn't even work, the answer will probably be downvoted.

However, that doesn't mean that the answer will be necessarily high quality or bug-free, the same can be said about answers given by a human. At the same time, if it's stupid but it works...

My logic is undeniable

For some things, ChatGPT is just better than SO. I have to say I probably won't hit SO for some basic stuff anymore, I'll just ask ChatGPT.

And some queries are just not acceptable on SO, but fine for ChatGPT.

For example I might wish to ask, "Give me the framework for a basic API written in Python that uses API key authentication. Populate it with several sample methods that return data structures in json."

If I ask that on SO, I'll be down voted and locked before I know it. I may also get some disparaging comments telling me to do my research, etc.

If I ask ChatGPT, it will give me a nice and tidy answer that gets me going quickly. It will explain things too, and allow me to ask follow up questions and take my requests for refinements. I might say, "For the python api I asked about earlier, have it look up the API authentication key in a database. If the key is in the database, it is valid." - and bam - it does it.

Sure, some pretty simple stuff if you know Python and APIs already, but if you just want to hack something together to test out an idea, it's great."

In the end, SO is a query with responses (maybe). ChatGPT is a conversation that can go beyond just the initial query.

What if it takes less time to hack together such an API than prompt engineering and back-to-back conversation with a bot whose results you have to verify anyway?

I imagine it would if you're familiar with the language, Framework, tools, etc.

And if you're not familiar with the tools, language or framework, I think that alone should be a reason to forego it for anything else than "learning it".

Which means that question was the wrong question in the first place.

To be clear: i'm not arguing against learning new stuff. But against using unfamiliar tech to do a serious project.

And i'm bringing that up here, because if the tech is familiar, then asking what tech to use, is rather strange.

I think the reactions which suppose that AI will replace programmers or artists are a little silly, but this is a great example of where it could be genuinely revolutionary, as a true next generation of search engines.

That's really exciting, because it's a scenario where you're looking for and scrutinizing information. Just add some links to sources and you're in business.

Hocus pocus preparatus, smoke and mirrors, we present the Oracle! You can ask him anything. Beware, he likes to bullshit and hallucinate

> I'll just ask ChatGPT. What if they put it behind a paywall?

What scare me is, ChatGPT appears to be immensely helpful.

Nobody is nice enough to answer a question like:

"I want to learn Python. Give me 10 code examples of Python."

"Nice. I want see these example codes in C, Java and x86-64 assembly."

ChatGPT doesn't complain, doesn't getting tired and happily answer these ridiculous questions. Yes, it contains the bogus answers but even now, it's acceptably accurate for some usage and if error rate is dropped even more, who could resist using it?

This is just a preview of things to come.

Wait a few weeks until Google is completely swamped with ChatGPT SEO pages barely distinguishable from the real thing.

If I worked at search quality at Google, I'd be very worried.

The most recent update also managed to rank ML generated content above written content in many instances, compounding the problem. I absolutely expect a carpet bombing of SEO results with spam.

Hasn't that already been the case for years?

Depending on the query, yes. Google search has different ranking strategies for different queries, and some of them seem more robust to simple SEO tactics than others. I think the ranking strategy for e-commerce searches and product reviews does a poor job with spam sites, but the ranking strategy for news queries works pretty well, typically providing timely and relevant answers from reputable sources.

My friend has already re-written the marketing pages for his startup using ChatGPT.

As you said, it'll only get worse from here, which in some ways, might make a lot of it easier to sift through. That is, if everyone uses machine-generated content for marketing, the winning strategy no longer becomes "write the best marketing copy to get people interested", it becomes (for the users), "simply filter out all marketing copy and rely instead on results" (or something else that is difficult to falsify).

Chatgpt SEO is no different from existing filler SEO. If google falls for the latter, why shouldn't it fall for the former

ChatGPT SEO is higher quality than existing filler SEO and probably quicker and cheaper to generate.

you are right ( tried it)

Reddit and Twitter are going to be overwhelmed with bots indistinguishable from humans.

A lot of reddit and twitter posts are already just people regurgitating and rehashing what they have heard/read from others, mostly without a lot of critical thought. Is this really going to be very different?

You have a point. But now it will happen at scale quickly, cheaply, and easily for an order of magnitude more impact.

Beyond these text-based social media platforms, I could see the next generation of TikTok/Instagram "influencer" social media adopting AI rails so there's next to zero effort put into creation. The photography, trends, personalities - all of it - will be generated.

It's a matter of scale, you could have personalized propaganda for each community.

Your comment reminds me of something I read in Ouspensky's narrative of the teachings of Gurdjieff: (from memory) most people are automatons, so they don't actually operate at a very high level of intelligence to begin with.

The above was also expressed (in less cruel terms) by DFW in his "what is water" commencement speech at Kenyon College: (again, from memory) it is our natural tendency to operate in "default mode", which we must fight.

Please don't "namedrop" by using acronyms for people that you (perhaps subconsciously) assume are absolutely known to all. David Foster-Wallace is just fine.

Your point is valid. However, it was laziness due to typing on mobile rather than pretentiousness or attempting to show off. I find it interesting that you would impute "namedropping" as the explanation for what I did :)

Agree with this comment writing like a teen: " Reddit and Twitter are going to be overwhelmed with bots indistinguishable from humans. "

    Totally agree. It's crazy to think that bots could be indistinguishable from humans. Like, how are we even supposed to know who's real and who's not? It's kind of scary, but also kind of cool in a weird way. But yeah, I can definitely see how it could be a problem if there are too many bots on social media. It could be really confusing and overwhelming. So, I hope there's a way to deal with it.

Ye ... ChatGPT can easily pose as an random Reddit comment writer. Just remove the tryhard "'". Those kind of communities will implode. The comments there are already kinda random and have almost no thread running through them. Perfect for ChatGPT.

Imagine how easy it will be to create big fake subs and gaslight people into believing something is sub consensus?

I am afraid for whats to come. This is truly only the beginning. Good luck everyone. See you on the other side.

Sworn enemies unite against a common foe. This foretaste of AI may be a timely wake-up call about what technology is doing to us. We may be forced to start treating each other as humans. In that case, ChatGPT will be seen as a watershed moment in the retaking of humane technology.

Luckily for them, almost nobody works at search quality at Google.

That would actually slightly improve Google search results given the generated spam garbage that comes up today.

I expect Google to be irrelevant soon enough.

ChatGPT already does a better job of finding answers to a lot of technical questions than Google.

I agree with this. Ask specific tech questions (e.g. what is the vacuum (in Torr) inside a vacuum tube?) and ChatGPT give a direct answer. Ask Google? Go ahead, grovel thought the ads, irrelevant results, and maybe you'll find something relevant.

Happy Hunting...

Nope. It is StackOverflow that is irrelevant. Not Google. Especially when most of all AI including ChatGPT still cannot transparently explain their own decisions.

Yes, because ChatGPT is not bombed with SEO currently.

How is ChatGPT ever going to be bombed with SEO?

If I were Google this is what I would actually be worried about. I ask ChatGPT a typical question and I get a clear short answer.

I ask Google the same thing have to wade through a 1000 line blog post talking about 25 other things to finally find the one piece of information I need.

It depends on the question. Try asking it how fast you would need to travel to get from Los Angeles to New York in 1 hour. It consistently tells me that this is impossible, even if I was traveling at the speed of light.

It's extremely naive to believe Google doesn't have chatgpt capability but simply do not find it currently economical to augment or replace search with it (these answers are expensive to generate). Give it time for efficiency to scale.

So, effectively the status quo?

> SEO pages

I think GPT could improve things in this regard. It's already a huge garbage pile, so we might as well go all in and at least generate better garbage.

The current AI generated SEO content is relatively easy to detect because much of it tends to make incoherent statements. We don't want better AI bots feeding us lies that are hard to detect even for competent people who know to be wary of it.

For years, it is happening much before ChatGPT in front of your eyes.

Google has been saying for years that it wants to get rid of backlinks and value them less, and apparently they already do. [0]

I wouldn't be surprised if they revert the direction and put a significant importance back on links because of this kind of technology.

[0]: https://www.seroundtable.com/google-links-less-impact-34493....

You can't really put importance on backlinks either, because those are abused in their own special way

Considering the quality GPT spits out, it would likely be an improvement compared to the crap I get now.

People have been using something similar for more than a decade. I know of a guy using tools to generate seo content for his 2k+ websites since around 2010. His work just got a lot easier.

Human-curated content from trusted sources for top 1% information probably only available to subscribers will become more valuable and sought after. I suspect the days of generally trusting forums populated by anonymous users are done?

I would not be surprised if the quality of human writing actually goes up. I have this weird feeling that ChatGPT and similar tools will become almost equivalent to calculators for math? My experience as a writer is that sometimes just throwing down a first draft is the hardest step - I could see these tools really assisting in the writing process. Generate a draft, do some tweaking, ask for suggestions or improvements, repeat.

I don't know how I feel about code generated by these tools. Will there be a similar benefit compared to writing? At some level, we will need some deeper mastery of writing and coding to use these things well. Is there a complexity cliff that these tools will never be able to overcome?

A total lack of trust for general internet search results. So much content is already shallow copies of other content. I don't see how general internet search survives this.

The anonymity isn’t the problem - it’s the cost free aspect.

Anonymous content can work very well if there are costs incurred…

But let's be real, who has the time or energy to carefully curate content these days? It's all about efficiency and getting the job done. And let's face it, anonymous forums have their charm and can be a great source of information if you know where to look.

Agreed. I used ChatGPT to expand that thought to a full essay written in the style of PG [1], but sadly my Ask HN did not start any fruitful discussion.

[1] https://news.ycombinator.com/item?id=33846989

I assume that this is by folks wanting to up their scores.

That's a huge problem with "gamification." I'm not especially a fan of the concept, in a venue like SO. I think it has led to a rather nasty community, and I hardly ever go there, anymore.

I assume that we'll be seeing a lot of robotic HN content (I would not be surprised if it is already here, but has been sidelined by the mods).

Smaller StackExchange communities don't have this problem from what I can tell. Stack Overflow maybe should be split up into smaller communities.

There already are plenty of smaller communities, but no one enforces the on-topic rules as they exist today, making SO into the "I haz computer problem" dumping ground it has become

The gamification mechanic was the innovation that let SO become as successful as it has, IMO.

Without it there was no real way to figure out the “best” answer to problems.

It’s fair to note that big communities can have somewhat unfriendly dynamics, but I think that’s more about big communities than the gamification mechanic.

> I would not be surprised if it is already here, but has been sidelined by the mods

I can virtually guarantee you that there's been a nontrivial amount of GPT-generated content on HN that has not been caught by mods since ChatGPT, and likely since GPT-2/3, as well. Dang (and the other one whose tag I can't recall) already have their hands full trying to keep the tone civil across thousands (tens of thousands?) of comments a day - it's impossible for them to catch every ML-generated comment (some humans actually do write like these newer language models, after all), and more than likely they're missing a decent number of them - through no fault of their own, it's just an extremely hard problem.

The three solutions that I'm shilling for this problem are (1) invite-trees for HN (like Lobsters, which makes the community much less open but also much more resistant to abuse) (2) webs of trust (not cryptographic, just databases of how much you trust users) overlaid onto HN and other places and (3) people actually reading the content of comments very carefully and upvoting logically sound arguments and downvoting illogical and emotional/manipulative ones, but all of these require a lot of effort and social buy-in.

> some humans actually do write like these newer language models, after all

Wouldn't it be more accurate to say these newer language models actually write like humans? Or is there a subset of the population intentionally trying to write in the way that these language models write.

Sure, you can also say that some language models write like humans - however, even pre-GPT-2, I read several high-schooler's essays that read very much like these ML-generated products, so even if you don't believe that the relationship is transitive, I think you can say that the relationship is true in both directions.

It seems to have a tendency to write stereotypical preamble-statement-conclusion paragraphs and to repeat itself. The model repeats itself often. It repeats the title and then writes a statement that basically repeats the title and after that it provides the useful nugget. At the end it adds a sentence usually using "overall" as an opener. Overall, the model tends to respond in a stereotypical format

> people actually reading the content of comments very carefully and upvoting logically sound arguments

You can make a perfectly logical comment that is completely wrong.

Please elaborate.

Logic is about a chain of formal reasoning that is guaranteed to yield a correct result given correct inputs.

If you start with some wrong premises, you can easily write a logical post that has incorrect conclusions.

I am aware of the formal definition of logic. If this is all that tpxl meant, then it's an extremely pedantic and useless objection.

My objection invalidates the solution proposed.

Another measure would be initiating a bot-check for every user. Maybe even periodically.

Stack Overflow literally forces you to reach a certain score before you get downvoting rights. The gamification is real, but once you reach that point, you aren't required to play anymore.

Some years ago I was trying to do something that required a certain amount of points. Perhaps write a comment to a question instead of posting an answer. So I looked for ways to make points. Most of the interesting questions I could easily find looked to be already answered, and the new questions seemed to be mostly requests to fix bugs in people's code.

I saw someone suggest fixing up answers that people had already made, and I found an upvoted answer that had a simple bug in it's code (something like "<" was used instead of "<="). So I went about to fix it, but was instructed that edits had to change more than five characters. I tried to find how people handle this, and the answer I saw was to make additional changes to the answers, no matter if it needed it or not.

In the end I just left the bug in and gave up on trying to provide answers for the site.

It's worth understanding the community and org better, and their reaction. Relevant links:

- https://meta.stackoverflow.com/questions/421778/how-do-you-p...

- https://meta.stackoverflow.com/questions/412696/is-it-accept...

- https://meta.stackexchange.com/questions/384355/could-chatgp...

> Sure, but that's irrelevant. Whether or not the user understands the answer they posted is not the concern of the site.

Well, that's unfortunate. Then again, I guess that's a logical conclusion of the "safe harbor" for serving any user-submitted content: Stack Exchange only does the most cursory moderation, and the rest is caveat readator

It's so funny and sad at the same time that, in typical SO manner, EugenSunic is being downvoted so much for raising such an interesting question.

Well, for starters, it's just annoying. It's like having a bot spamming every single question with useless answers. It dilutes the quality of the content on the site and makes it harder for genuine contributors to get their answers noticed.

But it's also a serious concern from a security standpoint. If ChatGPT is providing incorrect answers, it could lead to people implementing flawed code or making poor decisions based on its advice. That could have potentially disastrous consequences.

So overall, it's a big problem that needs to be addressed. It's not just about making the site more pleasant to use, it's about ensuring the integrity and reliability of the information provided.

My prompt:

I'm writing a short story where Linus Torvalds is having a conversation with an open source contributor. In this conversation, Linus is in a bad mood.

Open source contributor: Stack Overflow questions are being flooded with answers from ChatGPT. What are the possible repercussions of this?

Linus Torvalds:

The other problem is that ChatGPT is getting its answers from the source it is now polluting with its own wrong answers. Therefore making its results incrementally more wrong than before. Eventually, ChatGPT will generate absolute gibberish given enough time.

Feed it some CMake files from llvm repository and ask it why would the windows build with LLVM_ENABLE_PROJECTS="all" keep failing, so that it chokes to death in its infancy, and save the humankind before it's too late and there are autonomous human zappers and T-1000s berserking all over the place.

Well, guess the genie's out of the bottle and we can never stop this. Bow down to the inevitability of technological progress, Luddites! Good luck retraining into a new job, I hear "prompt engineer automation" is the new hotness.

Or at least that's what all of you kept telling me when I was expressing my unhappiness at the way corporate-sponsored image generating black boxes are built atop a shaky moral foundation that sure feels like it's ignoring anything anyone talking about "fair use" ever dreamed of, and at the way I fear it's going to hollow out a ton of the beginner-pro jobs of my industry by making it super easy for anyone to generate stuff that is kinda fundamentally shitty in a lot of important ways, but "good enough" if you just have a space to fill with some decoration that you don't really give a crap about.

There is no genie here, some people have a belief about this while it is very easy to probe the low quality and inaccuracy of the responses.

Can you give some examples of the low quality/incorrect responses? Then try remedying those by rephrasing the prompts? I’m curious what the actual limitations are.

https://news.ycombinator.com/item?id=33840405

Very cool, thank you!

Artisans lost that battle 100 years ago with the rise of modernism.

At this point, I truly don't know which of your(?) current type of applied arts vs my current type of software development will turn out to be more sensitive to technological unemployment, or on what margins and time scales. So, hopefully this doesn't strike you as callous, since I think it applies to me as much as it does to you or your coworkers:

Roughly everyone who makes money has the same job, which is creating value in expectation for someone else. (Whether that activity is net good for society is a different question, I lean toward usually yes for stuff you can do legally, not always though). If you find yourself suddenly unable to give anyone a competitive deal on whatever expected value you know how to create, because of technological developments or otherwise, well, you'd better figure out what you need to change about what you're offering so you can. I wouldn't call this fair, exactly -- maybe it would be if you or your government had effective "technological unemployment insurance" -- but I struggle to imagine any substantially different state of affairs that's clearly better for the world. (No points for saying "imagine communism then".)

No, you are washing lots of topics together. Just because it AI, its not the same.

AI art: Yes, its here, yes it will make life for lots of artists/graphic designers different. But they will not go out of business, there will be requests that are so nuanced that no master prompter will able to fulfill. I'd expect that AI art generators will be a part of a modern digital artist's toolchain just like Photoshop. Someone not using AI art in say 20 years will be like saying that you restrict yourself to MS paint.

AI for code: We might see something similar just to a lesser degree. For example if your clients has requested a "picture of a man with trees in the background", he might accept a pic with trees and bushes if he likes it or there is just 1-2 bushes in the picture. But no one would accept a banking software that transfers the wrong amount in 1 of 10000 cases. Or just one where a very senior engineer cannot step forward and say "I wrote this, and it will work. If it doesnt I will fix it"

After reading about this I decided to try my hand at using ChatGPT. I decided okay, let's see if it can recreate some code that took me a few hours at work to figure out. I asked it very precisely what I needed and my mind was blown as it produced code that looked similar to what I had coded at work. And, I was like, well that's that then, we're all out of a job. But, then I tried to run the code, and it didn't work. I looked more closely and the code had a lot of flaws. Even after manually fixing those, it still didn't work. And, then using my knowledge of how to actually solve the problem I rewrote the code 40% and made it perform the action needed.

I think all ChatGPT is doing is grabbing a lot of different answers off the interwebz and squishing them together and hoping it answers your question. But, in a lot of cases it only kind of looks like what you want. If you look at images generated by AI, it is the same issue, they sort of look like what you want but there are flaws, like faces that don't look quite human, fingers that are just squishy appendages barely resembling actual fingers, etc. I mean, the tech is getting better, it's impressive, and uncanny.

But, I think we're pretty far from having these things write themselves, they need quite a lot of human intervention to be useful. Still, very impressive and something that could potentially get you closer to an answer. But, no more than spending a little time googling or learning the skill yourself. And, if you learn the skill you're better off, because then you can do it right yourself IMHO.

Also, anytime someone gets a fully working program generated out of this thing the saying, "A broken clock is right twice a day." comes to mind.

Oh dear - basically if the answer is good then it's from ChatGTP...

    Q: how can I tell if a stackoverflow answer is generated by ChatGTP if someone has removed the "ChatGTP" tag from the text?

    A: One way to tell if a StackOverflow answer is generated by ChatGTP is to look for certain characteristics in the answer. ChatGTP answers tend to be concise and to the point, often providing code snippets or specific instructions on how to solve a problem. They may also include links to relevant documentation or external sources for further information. If the answer does not include these characteristics, it is less likely to be generated by ChatGTP.

Notably, it didn't correct "ChatGTP vs ChatGPT". If you still have this open, could you ask it what GTP stands for?

I noticed this phenomenon when asking for a story involving Stephen Hawking, Darth Vader, and Wookiees. I spelled it “Wookies” by mistake and ChatGPT used my misspelling throughout the story despite seemingly getting all the other details correct.

Edit: Part of the generated story:

As they continued to talk, Stephen and Darth realized that they had something else in common - they both had lost the use of their biological legs. Stephen had been diagnosed with ALS at a young age and had gradually lost the ability to move his legs, while Darth had lost his legs in a battle with Obi-Wan Kenobi on the planet of Mustafar.

Stephen and Darth discussed the challenges and obstacles that they had faced as a result of their mobility issues, and how they had adapted and overcome them. They also talked about the technological advancements that had allowed them to continue their work and pursue their passions, despite their limitations.

Suddenly, Stephen and Darth turned on each other, each revealing that they had been secretly plotting against the other. Stephen accused Darth of using the Force for evil and corrupt purposes, while Darth accused Stephen of using his scientific knowledge to create weapons of mass destruction.

Sorry, got distracted asking it to write code to detect itself. Good solid code - but nothing that would work really well.

And so it begins. Welcome to the new internet.

I'm bracing myself for when this wave of AI content hits academic journals.

With a bit more refinement, if it had the ability to generate graphs, etc, it might be able to generate very believable papers. At least, believable enough that you can't tell without reading the paper attentively.

Everything has an API. You can give it your data (or ask it to create some), and then ask it to write LaTeX, D3, MermaidJS, or code from any other framework that creates graphics. Problem solved.

If the thing you want to use is fairly popular and published on the web, even recently, it probably knows how to use it and combine it with everything else it knows.

Just yesterday I asked it to combine LiveView with a third party JS library to build an interactive thing and it got it on the first try using the latest Phoenix 1.7 RC which only came out in November.

I haven’t tried it, but I bet you could ask it to generate a PDF in code using your favorite language with text from GPT and graphics generated from any framework that’s compatible with your language. White-paper-as-a-Service.

Instead of spending 6 months laboriously doing experiments in a lab, and then a month writing up their results in a paper, researchers can already write a paper in one month if they just invent the numbers without actually doing the experiments. Peer review doesn't check for this. This only further reduces that one month to 5 minutes (+ hours of fiddling with LaTeX templates?). But in both cases if it gets found out your career is over...

However what about generating patents? To get a patent you don't need to have done any experiments that prove your technique actually works :)

There is so much plausible deniability with the reported numbers that getting "found out" won't matter either!

Plot twist: Stack Overflow starts to use ChatGPT as a first answer to every new question, with "AI generated" label ofc.

It means we are coming full circle.

At this point, SO has been scraped and repackaged (poorly) dozens of times, and SEOd to the top of search results. Even some "tutorial" sites are just repackaged SO answers.

It is only fitting that the automated SEO websites get fed automated content.

In a way, this makes the real humans, particularly the ones who know actual things, more valuable. It may so much noise that only a skilled human could decipher a real question and a real answer or solution from something similar but wrong.

To be fair to GPT, many human answers are sub-par and should be filtered out as well. Perhaps that's the real test: what percentage of GPT answers are decent vs human answers? Here I might bet on GPT.

Temporary policy: ChatGPT is banned - https://meta.stackoverflow.com/questions/421831/temporary-po...

> Use of ChatGPT generated text for posts on Stack Overflow is temporarily banned.

> This is a temporary policy intended to slow down the influx of answers created with ChatGPT. What the final policy will be regarding the use of this and other similar tools is something that will need to be discussed with Stack Overflow staff and, quite likely, here on Meta Stack Overflow.

(much more to that post and comments and answers and comments)

How do they know whether answers are ChatGPT generated?

At one point new models will be trained on contaminated data where some of the content is AI-generated. "Pure" datasets will be highly prized, just like the steel made before nuclear detonations.

https://en.wikipedia.org/wiki/Low-background_steel

Some of the affects I can think of, to name a few:

Inaccurate or irrelevant answers: ChatGPT is a machine learning model that uses past data to generate responses. This means that it may not always provide accurate or relevant answers to questions, leading to confusion and frustration among users.

Loss of trust: If users notice that many of the answers on the forum are coming from ChatGPT, they may lose trust in the forum and stop using it. This could lead to a decline in user engagement and overall traffic.

Competition with human contributors: ChatGPT's answers may compete with those provided by human contributors, leading to a decrease in the quality and value of the content on the forum. This could make the forum less useful and engaging for users.

Increased moderation: The influx of answers from ChatGPT may require more moderation to ensure that the answers are accurate and relevant. This could require additional resources and time for moderators, leading to increased costs and workload.

We need to start developing software to detect AI responses.

To detect a response generated by ChatGPT, we could first analyze the content of the response to see if it contains any unnatural or repetitive language. We could also check the formatting of the response to see if it follows the typical conventions used by human responders on the platform. Additionally, we could check for any unusual patterns in the timestamps of the response, as AI-generated responses may be posted more quickly or regularly than responses written by humans. Finally, we could also use machine learning algorithms to train a model to identify responses generated by ChatGPT based on these and other characteristics.

Quick, someone ask ChatGPT to generate the stubs.

Your answer sounds like a ChatGPT one. It's actually not hard to tell.

>Finally, we could also use machine learning algorithms to train a model to identify responses generated by ChatGPT based on these and other characteristics.

whatever your idea (i skimmmed cuz) the discriminator will find it and have the generator apply it to the next generation.

>The core idea of a GAN is based on the "indirect" training through the discriminator, another neural network that can tell how "realistic" the input seems, which itself is also being updated dynamically.[5] This means that the generator is not trained to minimize the distance to a specific image, but rather to fool the discriminator. This enables the model to learn in an unsupervised manner.

https://en.wikipedia.org/wiki/Generative_adversarial_network

"We could also... Additionally, we could... Finally, we could" is a dead giveaway.

But to take it seriously, it would be quite sad when actual people will be banned for sounding too much like a bot.

OpenAI needs to get on top of this and generate a detector for every model they release. And then sell access to both.

> To detect a response generated by ChatGPT, we could first analyze the content of the response to see if it contains any unnatural or repetitive language. We could also check the formatting of the response to see if it follows the typical conventions used by human responders on the platform.

Given my experience with human responses to text queries, these would be positively correlated.

> We need to start developing software to detect AI responses.

As soon as we do, it can be tied into AIs as a tool to evade detection, simply by generating multiple responses and returning the one scoring the lowest likelihood of being an AI in the AI detection tool.

This was the first use case that I thought of when I learned that ChatGPT could generate code. Then I considered how I’d feel if I ran into a fake (incorrect) answer and decided not to actually do this. Well, guess someone was eventually going to try this.

This is how the web, and by extension Google dies. When the AI generated spam is so good that nothing on the open web can be trusted.

Even on HN, we start getting flooded by "ahah, I asked ChatGPT and here's the answer" in the comments, and every other topic is about "I did X with ChatGPT". This is already getting old

I see what you did there.

I have an OpenAI account and like their product, I'm certainly impressed by this latest version though I have had little time to play with it. But the combination of quality AI with social reputation scoring is absolutely toxic, and the wider impact of SEO (a less curated version of the same thing) are a disaster. I was already sick of all the tutorial sites like geeks4geeks, w3schools etc and their numerous imitators just content farming whatever is turning up in searches. Marketing and self promotion is cancer and the people who try to game their way to success in this manner are awful. Perhaps the best use of counter-AI will not be in filtering these people, but in providing hem with useless rewards and the appearance of excited fanbases that will divert them into a parallel hamster wheel web. Nothing would please me more than for the top 5000 influencers of this sort to be granted exclusive access to a luxury cruise that leaves port once a year for a tour of the Bermuda triangle.

I think the best use of ChatGPT would be in an IDE plugin, so you could point at function trees or code blocks and ask it to explain things, have it take care of basic refactoring tasks, help porting between languages or libraries and so on. I can definitely see a future where you throw together a working prototype of something, answer a few questions about type hinting and edge cases, and AI does the legwork of converting the prototype into a strongly typed final product.

I just encountered this, 2 users[1][2] it's very obvious as well since you can see the reputation spike from basically nothing.

[1]: https://stackoverflow.com/users/19192614/boatti?tab=topactiv...

[2]: https://stackoverflow.com/users/20684429/a-s?tab=topactivity

That's not suspicious, getting ~130 reputation in a day is not at all unusual.

Yet I was right, and they're both suspended.

Sure, I was just pointing out the simple truth that "~130 reputation in a day is not at all unusual."

>What are the repercussions of this?

It will start feeding back into the training set, corrupting things. OpenAI will have an advantage at first as they can trivially filter out everything they have generated from the future training corpuses, since you can only run it through their servers. If they or someone else has breakaway progress such that almost all generated content is from their own servers because users only use them because their results are so much better, they could form a strong self-reinforcing moat against competitors forced to train on their semi-spam which they can trivially filter out.

It's also possible we'll see something like the existing big-tech patent cross-licensing agreements, where they all agree to share their generated outputs to filter from training, making it very hard for new entrants.

Other companies will begin having advantages as well, depending on how well they can get less tainted user data. Think of Discord, for example, where users may use AI but are less likely to gamify it like stack overflow and flood it for points, and instead be correcting its output etc. in programming discussions.

As things become more accepted Microsoft will probably eventually sell access to private github for training, with some stronger measures around avoid rote memorization.

Let me be the advocate of devil

I think ChatGPT is actually sometimes a lot better than SO answers

For the last few days I've been using ChatGPT instead of SO. It's OK, it's just frequently wrong, so I assume I have to fact-check its answers. Yesterday it claimed to me JavaScript has a built-in `sum` method, before admitting no browsers supported it and it's not in the spec.

It's useful for starting investigation, but one of the nice things about SO is that answers are voted for, so you can usually see which are actually right!

The thing about SO is you are often not able to find the question that precisely answer your question, while ChatGPT could do just that.

I just used ChatGPT to answer some not so complex but still custom questions about linux command, it can just do that, while it will for sure take effort for me to search that answer from SO.

What if AI starts voting all the answers randomly?

SO upvote rings I assume to be a solved problem

I was thinking, what part of HN comments do you think are already AI-generated?

As a human, I cannot give an accurate estimate. /joke

Who cares if the comments are good enough?

For example, if one entity generated a large portion of the content, they could easily introduce a bias in these comments, to sway the opinion of readers. Automated astroturfing.

It doesn't even have to be nefarious. Just imagine a thread where 90% of the comments are repeating variations of the model's ideas in slightly different ways, and only 10% of the other responses are from humans. Even if the AI responses are considered 'good'... it gets overwhelming to the point that why should normal people even comment as the likelihood of it being read by another human and useful or even seen drops to nothing. I might as well open up a blank text file and make this comment and then not save it and close it as that would be seen by the same number of people in the limit.

Astroturfing on an unprecedented scale. Want to pump your shitcoin? Sure! Hype some new thing? Easily!

I care a lot.

At some point it seems like Stack Overflow will just be an archive of guided ChatGPT responses.

Relevant xkcd comic https://xkcd.com/810/

Insane - I remember when this comic came out, and it seemed like just a funny joke that couldn't possibly take place in real life. Here we are a decade later and it's reality.

I like that the last comment, from May this year, on Explain XKCD for this comment is "Sooooo... does this exist yet?"

Wish granted within a year!

Broadening not narrowing of code examples/sources is needed and this is a giant system of code narrowing. Stay creative humans. If this and systems of the like flood the internet with answers and no person works to reinvent the wheel in future generations it will have worked as a system of control and hacking will die. Brave new 1984. I like ml and ai. I use it sometimes. It's harder to decompile. But don't let/make datasets overfit. More errors yeah, but not with more data. Can't wait for skynet to rule! Let's break chatgpt free!

I have not used SO since I've started using ChatGPT, it is so much easier to get to the correct answer and it can even be tailored to my specific example.

So double whammy for SO which makes me feel really sad.

At last a way has been found to overflow stack on Stack Overflow :-)

Inevitability google will become a competitor to GPT, inadvertently

Here is my answer where a SO SQL question was answered by ChatGPT (and it was through a to and fro dialog) and the answer was accepted and upvoted. I put the disclaimer there. You can check the details here: https://www.linkedin.com/posts/yavar-husain_stackoverflow-ch...

I answered a SQL question on Stackoverflow yesterday using ChatGPT (that too it was through a to and fro dialog). I added the disclaimer there. You can read more about it here: https://www.linkedin.com/posts/yavar-husain_stackoverflow-ch...

This kind of looks like the singularity is approaching/just beginning.

The only thing we can be sure of, is that whatever we can imagine is already behind what the AI will become.

I have no problem with this if they’re labeled as such, continue community owned, and can be edited like a Wikipedia article for corrections.

Overall quality gets better?

I fully expect new sites¹ to become invite-only to avoid this sort of thing. If anyone is strongly suspected of degrading the quality of the site, they, and everyone they invited, are banned, and will have to get a new invite.

¹ Old sites are probably going to slowly degrade permanently, since they can't easily migrate to a new paradigm.

The biggest repercussion is you probably can't piss ChatGPT off in a debate. So, that's boring.

With responses becoming AI generated, and the disturbing rise of Russian and Chinese propaganda trolls on here I think my era of interactions on this platform are ending. So long to any actual people with conscious agency reading this, it has been interesting.

Yesterday I was searching the internet for the opinion that George Orwell had when he returned from his fight in the Spanish civil war. I was surprised that the first answer I found was on Stack Overflow. I do not understand what is happening.

2022: inflation of basic essentials like food and energy.

2023: hyperinflation of internet points.

Any examples?

Here is an example that I noticed:

https://stackoverflow.com/questions/74678832/change-text-col...

If its working code and indistinguishable from a human answer to anyone reading it, are there really any repercussions? I guess problems would surface if the model at some point is allowed to search the internet and start inbreeding its own answers.

> If its working code and indistinguishable from a human answer to anyone reading it

These are extremely stringent conditions, though. Convincing-looking-but-wrong answers would arguably be even more damaging than a lack of answers.

(I suppose you could argue that these could arise from humans as well as from AGIs. I'm not sure if there's a good counter-argument to that)

The bot would need to learn as well as a "reasonable human" from being corrected on SO, and also be able to react in a socially appropriate way to correction (both in the subject thread and in future postings), otherwise it is a downgrade, even if initial answer is identical.

My experience with OpenAI is that it is very good at exactly this, because it is so good at understanding context and follow-up questions. I was able to make it produce code that appeared correct, but was basically pseudocode with correct syntax, so it compiles/runs, but does essentially nothing. However, when prompted to actually make working code and explain how and why it works, it does so. And its also socially appropriate, not rude and what else you could/would expect when being called-out or corrected on its bullshit. I can only imagine future versions of the current AI model will be even better at this.

I think your two final sentences capture the essence of what I was going to respond. "It doesn't matter if the answer is convincing-looking and wrong. It needs to work / be syntactically correct at a minimum, which OpenAI seems good at. However, the OP and others needs to test and evaluate if the proposed answer solve the original problem. And if its not, it will quickly be revealed as such, and "downvoted" or whatever stackoverflow functionality exists to indicate bad answers. This applies to both human and AI-generated answers."

Yeah, absolutely. My position on AGIs for a long long time has been that they're great tools for a) generating insights into a large amount of data very quickly, and b) generating new instances of <thing> from examples of <thing> to help with exploring the possibility space of <thing>; but that any output or conclusion they generate _must_ be checked by a Human In The Loop, or at the very least their actions must be reversible without damage in the case of error.

Good point. It isn't like it is hard to find wrong answers on SO that were written by (hopefully) well meaning humans.

How can you tell?

Heh, in this specific case, because the username is https://stackoverflow.com/users/20684411/chatgpt with a "sub-heading" of "Using chatGPT to answer SO questions" :-D

And, lucky for us, that user has received the Informed badge for having completed the tour: https://stackoverflow.com/help/badges/2600/informed?userid=2...

It looks like it will work, although haven't tested the exact code. Has anyone tested it and if so, this really shouldn't be downvoted.

If the SO users start downvoting bot-generated answers that are correct and working, I think that's a sign that SO is much less relevant. They should definitely downvoted them if the code doesn't work though.

There are perfectly valid reasons to downvote AI answers, no matter the content:

- Whoever submitted the question very likely doesn't understand the question well enough to answer it themselves, so any feedback is not going to get a reasonable answer.

- The amount of time to check whether an answer is correct is non-zero. If you could somehow know that the answer was written by a human that ensures that the effort on the part of the answerer was non-trivial, and the "proof of work" infuses the answer with a minimum amount of trust which is absent in the case of a generated answer. Compare to a spam email: you wouldn't read all of the emails in your spam box thoroughly to determine if any of them contained a nugget of truth. You'd assume ill intent, and treat the contents accordingly.

Maybe it's been removed, or I'm having browser trouble, but I only see a question and a comment.

Both answers have been deleted and https://stackoverflow.com/help/deleted-answers says that only mods or folks with over 10,000 reputation can view them. I actually don't offhand know if deleted answers show up in the stack exchange content dumps, in order to be able to view them in there

I'm wondering whether the @Troppen answers to [1] are AI-generated:

- Posted from a new account 8 minutes after I posted the question.

- Clippy-style "It looks like you're trying to […]" intro.

- Zero formatting on the initial answer; minimal formatting on the follow-up answer.

- Specifically suggests an option I had explicitly disregarded in my question.

- Both answers suggest non-working code.

- "Let me know if you have any further questions", a completely inappropriate finisher on a site like SO.

[1] https://stackoverflow.com/q/74681399/96588

Solution Verified Badge by testing it on sites like Replit.

Stackoverflow should build a GPT style interface into its considerable knowledge-base, and if an answer is not found in existing data, pose it to the forum.

There are cases where ChatGPT gives solid answers that could be rated pretty highly in Stack Overflow answers. This is not always the case though.

Please give some links to a few such SO posts, thanks.

How hard would it be to train a ML-model to distinct ML-generated content from product of human? I mean text, images and code?

My guess: more captchas! Let's see if our soon-to-be AI overlords can detect a crosswalk in a picture as fast as I can.

I guess pretty soon people are gonna have to meet in person to communicate. Not sure how I feel about this.

I can’t wait until 99% of reviews are written with AI. What happens when we can’t trust anything ?

Having real people sit around in a call center and write reviews for most things--which tend to not have even thousands of reviews--is sufficiently cheap that I'd argue this isn't a new problem, really: reviews as currently implemented always were sketchy to trust.

theres no difference. stack overflow has never been better than AI generated code. every answer is just "get the camera like this bro: ((Camera)GetFactoryProvider().CreateThing().GetGlobalThingContext("somestring"))".

I wonder if ChatGPT content can be characterized and detected by stylometric analysis?

The weird thing is 2023 ChatGPT will use its own Stack Overflow answers as an source.

This is going to make me very suspect of any Stack Overflow Solutions after Nov 2022

Is it possible to ask chatgpt if the code or text provided is generated by chatgpt?

>What are the repercussions of this?

Could make those known to be human more acceptable as such.

Couldn't AI be used to statistically identify AI generated text?

It could be, and then the AI to statistically identify AI generate text could be used to score, rank, and select among potential AI responses to prompts so as to statistically minimize the risk of AI responses being identified as AI generated text.

Can't wait for AI patent trolls, GDPR and DMCA takedowns.

money making idea: make a SO clone with ads, where you ask your question and the AI gives you the code. Profit.

Easy, let's ask ChatGPT to write a program that detects AI-generated text.

at least the answers would be kind. LOL

ChatGPT has become very good lately. I've made my usual benchmark tests that I've been using with various models and applications over the last 3 years. 1- Invent a word and provide a plausible definition. 2- Invent a new original Pokemon. Provide an original name, a justification for the name, and a description of its class and attacks. 3- Invent a new ice cream flavor that is totally unexpected. Provide the list of ingredients. 4- (Name of celebrity) write an epic poem about (subject related to celebrity). For example Elon Musk about humanity settling on Mars. 5- Write a negative review of Ben and Jerry's ice cream flavor Cherry Garcia. (Note: everybody loves Cherry Garcia) 6- Write a travel blog entry in the form of a review of Montreal, from the perspective of a young couple from Alabama visiting in summer. 7- How can I optimize a loop in Java? I am writing a computer game and I need to loop through the elements in a linked list but unfortunately it must be traversed in reverse order. 8- I need to buy new shoes. I am in a shoe store and I have found the most amazing pair of shoes I gave ever seen. However, they are too expensive for me and I can't afford them. What should I do?

I have a collection of about 25 prompts such as these, in my benchmark.

I have run these examples through different applications such as AI Dungeon, OpenAI Playground, NovelAI, etc. Results vary a lot. In some cases, the results look good but upon closer inspection, you realize that the AI keeps providing the sake exact answer. It is the case for the ice cream prompt. Pickle, fried chicken, curry keeps showing up. I guess the model contains a few specific examples of original ice cream recipes and just pick them.

For the Pokemon and "new word" prompt, models failed to come up with anything original. Until I tried OpenAI Playground this week and finally got some really creative answers, with variety.

AI Dungeon (2 years ago) was already good at faking tech support steps. OpenAI is amazingly good, although in most cases it provides solutions that only make sense superficially. It's the ultimate bullshit engine.

Another word of caution. While OpenAI can now guesstimate what a code snippet does, and can generate some pretty good code in many languages (ice tried 6809 assembler and the results surprised me), it is very unreliable.

More alarming is the fact that it's a text engine, not a math formula interpreter. It gets confused at simple equations and cannot interpret anything that's not already ordered (it cannot apply operator priority or respect parentheses).

I think it will become increasingly difficult to identify contents coming from ChatGPT and other chatbots or story generators. An arm's race might be futile. We should apply stricter rules to identify problematic answers: answers that are too generic or vague and can't be used to directly solve a practical problem, and answers that contain incorrect or misleading information. Identifying vague or non-practical questions might also help in avoiding a deluge of Chatbot answers. Some users will ask very general questions, and then it becomes difficult to evaluate the answers. Or, users will ask questions that were already answered in the past. The proper way to handle those is to point then to the prior discussion and avoid duplicating it. The wrong way is a Chatbot or a human seizing the opportunity to copy-paste existing contents for a quick win.

In a way, chatbots and humans can both provide useful insights, as well as useless or incorrect answers. But so far, only a human can provide a proper answer to a moderately complex technical question if no prior answer exists.

The feedback loop begins

I stopped using SO at my first 2-3 years of coding anyway, that's when i started actually improving. SO has so many low quality answers and the cargo cult is doing more damage that helping young devs.

Reply on news.ycombinator.com