Is Hacker News Bullshit?
liamp.substack.com
liamp.substack.com
GPT-3 is a red herring; the issue was the generic, baity title on a popular theme. Those routinely get upvotes because people see words like 'procrastination' or 'overthinking' and instantly think of their own experiences and ideas and want to talk about them. Such threads are not about the article, they're about the title, which the author admits writing ("I would write the title and introduction, add a photo"). Title plus introduction is already more than most people read, so this case is not what they say it is—which is consistent with their other misrepresentations, including the false claim "only one person noticed it was written by GPT-3".
Edit: as YeGoblynQueenne and minimaxir point out, given the other misrepresentations I'm not sure why we should believe that the article body was exactly "written by GPT-3" either, especially because the author describes it this way: "as unedited as possible"—in other words, edited. That doesn't much matter, though, because the salient inputs for the HN thread were the parts they wrote themselves.
We usually downweight such submissions, but occasionally we let one through because it seems healthy and interesting to have such discussions in the mix—just not too many. If there was a failure here it was that the moderators didn't look at the article. Or maybe we did; on my computer the title, photo, and introduction cover the entire first screen. Maybe that seemed good enough to let the discussion run.
You should assume all GPT-3 related text (including app demos and screenshots) is cherry-picked unless you can see the text generated in real time for yourself.
Really, this whole "OpenAI" is an oxymoron and I hope a truly open alternative will appear.
___________
There have certainly been signs of “voting rings” (or something else that causes similar end results) for over a year now.
It has long been obvious that downvoting gets abused, as mentioned in other comments here.
Does HN ever mean to acknowledge and address that?
As for downvoting, people have been saying all the same things for well over a decade. The discussion circulates endlessly. Does downvoting get abused? Sure, but evaluating the system by specific examples is like evaluating an immune system by looking at what a couple of white blood cells did. You don't reprogram your immune system based on that—especially because the examples that stand out are always the cases that the system didn't handle perfectly.
Yet HN clearly did reprogram their system, by requiring a minimum karma before an account is able to use the downvote button, and not allowing users to downvote replies to their comments, instead of just working like Reddit.
What was the reason for that?
Someone seems to have recognized the downsides of allowing everyone to bury comments, and made an attempt to limit that.
That limit is no longer effective, as often seen in the drive-by downvoting on divisive topics and the burial of facts if somebody doesn’t like them.
As the total number of HN users increases it becomes easier for all accounts to reach that minimum karma, e.g. by posting funny comments (HN is not immune to memes) and riding on bandwagons.
As I said, this discussion circulates endlessly; if you use HN Search you will find that all of these concerns/complaints go back to the beginning of the site. People have passionate, conflicting views about how downvoting should work, and commonly feel like the site is deteriorating because it doesn't work the way they would design it.
But what was the reason? What is it supposed to achieve or prevent?
> As I said, this discussion circulates endlessly
That sounds like a reason to address this, instead of ignoring it.
“People keep talking about it so let’s leave it the way it is”?
I'm here for the life-giving, messy breath of fresh air not found in most online spaces.
The reality is that moderation is about helping people tolerate people unlike themselves and about buffering the worst communication gaffs. But beneath it all, sometimes people are just grumpy or mean and there is no fixing that. You can only mitigate the worst expressions of it, not stamp it out.
Right now, there's a Pandemic on. The world generally is pretty cranky and wanting to insist on someone fixing Something so they can feel like their world isn't spinning out of control.
Members successfully bullying mods into fiddling with voting rules in a short-sighted manner to get some momentary sense of control in life isn't going to fix the pandemic. It's just going to undermine long-standing best practices for a well-run forum.
One must make design choices. I think HN's design choices in this area are good and I've not heard sufficiently good reasons to change them. Not everyone agrees, but everybody will never agree.
I find it hard to believe that in a discussion which supposedly “goes back to the beginning of the site” you haven’t heard sufficient reasons yet.
I’ve often thought about gathering a collection of comments that were perfectly reasonable, well-behaved and factual, but still grayed out because somebody didn’t like them. (It takes, what, just 2 or 3 downvotes to make a comment practically invisible.)
Perhaps enough such examples would be sufficient to recognize that there is a problem.
Btw, if a comment is faded and you're having trouble reading it, you can click on its timestamp to go to its page, and in that case it should be rendered normally.
- https://news.ycombinator.com/item?id=24040450
- https://news.ycombinator.com/item?id=24054863
- https://news.ycombinator.com/item?id=24049094
I guess there's a reason for it
EDIT: Confirmed by the OP (wporr) in a dead reply to this comment (which he just deleted), who is indeed shadowbanned from HN. Which makes this submission blatant ban evasion.
I thought this particular experiment was funny, (mostly) non-offensive and insightful.
Also 2 of the 22 top level comments mentioned GPT-3 and several others are clearly just riffing off the headline. I think the percentage of people who actually read the article and realized something was off about it was higher than the author is implying.
I don't blame them, the language is not pleasant to read, the ideas superficial and a cursory glance is enough to give a general idea of the content.
That said, the article was still surprisingly coherent for being auto-generated. I wonder how much was edited to make it so. If no or minimal editing was applied then this is darn impressive, and we just might indeed start seeing more texts like this. Or miss them, as it were.
Partially off topic but I personally use the FOSS android app "Materialistic" to access HN. It offers a much better comment navigation experience than ynewscombinator.
this is pretty hilarious. Although the genre has probably much to do with it, virtually all business/productivity/ lifestyle/self-help stuff has the same sort of generic tone and is full of crap
Articles are also upvoted because of the discussions in the threads so I'm not that surprised a lot of "warm body" type threads make it to the top.
Downvoting prevents others from seeing a comment and there are no safeguards to ensure that facts don’t get buried by bandwagons.
HN clearly means to control downvoting because it requires a minimum karma, but it’s no longer effective. Perhaps also limit the number of upvotes an account can dole out and a comment can receive...
The comment in question was insightful and displayed intuition and judgement a level up beyond other people. Really every other comment was a load of crap, and the one comment of actual substance was buried. Many comments like this are voted down on HN. These are the comments that should in a perfect world be the top voted comment under any topic.
The fact that this is common shows that there is a huge difference between what people actually value and what they think they value.
People value opinions they agree with, not opinions that are new, different or insightful.
What's even more scary is the fact that the comment wasn't insulting. It was factual and to the point, yet the replier accused the original commenter of punching below the belt and not being civil. Accusations that originated completely from his imagination.
It shows that people in general don't just respectfully disagree with differing opinions. People attack all opinions that aren't inline with theirs.
And what's even more crazy is no one is aware of this. That responder thinks he's being civil and logical when in reality he's being stupid, illogical and biased.
I know the policies of HN are done for the purpose of preventing HN from becoming like 4chan or reddit. BUt where do I find a community where that guy who made the insightful GPT-3 comment will get voted up?
Somewhere buried in every HN topic is a comment that is ahead of it's time but you won't see it because it's been buried to oblivion by the mob.
> It shows that people in general don't just respectfully disagree with differing opinions. People attack all opinions that aren't inline with theirs.
It's not just that. The achilles heel of HN's readership/voting behavior is its bias towards positivity and validation, and against criticism and sarcasm. I've got tons of fake internet points here, so I often do some experiments to test this out. As a general rule, with some exceptions: No matter what the topic is, if your comment is positive, shows approval or otherwise supportive of the topic, you'll tend to get upvotes. If you raise concerns, point out flaws, or express cynicism, you're going to get downvoted.
So when something comes up that's actually false or suspect, the correct skeptics get downvoted, and the incorrect people praising the article are promoted to the top. I wish I had access to the raw data to do a sentiment analysis on the comments and compare it to voting trends. I guarantee they are correlated!
Ironically, the first vote on this comment was down.
We need to switch our brains to focus on what is being said instead and a lot of people have a problem with this.
When there's no way to define the authority of who is saying it, the authority is transferred to the subject instead, in this case "Hacker News".
People will have a tendency to flock on the authority represented now by the subject.
(BTW, in my experience HN is the best place i know where "whats being said" is more important and have an authority on its own, and the reason why i keep coming back here)
I agree. Still, even though it is the "best" it is still quite bad and this very article proves the point.
And no, I don't "fear" or am "frightened", as some using dominating speech would have it. There are real concerns in the world that need to be openly addressed, and people need to demand accountability and fairness for those issues.
Clever "retro" hacks will always gain traction (building a widget out of old vacuum cleaners), anything socio-psychological will usually get traction (dating apps, blog posts about loneliness), anything on advanced math/physics/logic will usually get traction (probably because it makes people feel smart, even though it's not accessible to like 99%), anything tangential to philosophy will get traction (again, makes people feel smart).
I really do miss when HN was a bunch of scrappy entrepreneurs that tried to build stuff. With all that said, it's still the best community left on the internet. At least it tries to abide by the principle of charity, and flame wars/trolling is highly frowned upon. Dang and everyone else does a very admirable job of keeping everyone in line.
I think these posts are popular because they generate meaningful discussion. More often than not, I learn more from the comments than I do from the article itself, because people will talk about their own experiences, point to resources that helped them, etc.. I don't see this as a bad thing at all--it's one of the primary reasons I visit HN.
Personally, I couldn't care less about the entrepreneurship side of HN. But that's the beauty of sites like this--everyone upvotes what they're interested, and more than one topic can occupy the front page at once. You click on what you like to read, I'll click on what I like to read. It's great!
It's 'easy' if you know how to do it. Many people actively try to figure it out and can't readily pull it off.
It's a system. If you know how the system works, you can get somewhat reliable results. That's not necessarily the same as "gaming" the system.
These are all examples of bullshit?
Yup, count me as one of them. Hask me anything.
Alternative, charitable interpretation: their big stumbles offer the best, clearest lessons for technical owners and managers; stands to reason they get a lot of attention.
Person1: "The posted has no substance and is pure regurgitation, it was probably written by GPT-3"
Person2: "Your comment punches below the belt and isn't welcome in a community like this. Maybe you're new here."
YIKES.
Edit: Of course the real world is like this too but I think there's a meaningful difference in the degree.
You can disagree in whatever manner you want, as long as you are oblivious to the obvious social cues of the people around you.
Whereas on HN, there is a score to tell you that you've committed a social faux pas. It's more immediate, so it's a little easier to figure out.
I've often thought about ways to integrate the scoring feature into real life to help folks on the spectrum learn to communicate with neurotypicals. Haven't squirreled that one out yet.
In real life, you're not grayed-out for being contrary.
Tangentially, the lack of body language feedback is a major reason people are such assholes online. It doesn't exactly feel like you are talking to a real person and you aren't getting realtime feedback, so it is harder to adjust to the other person or even just treat them reasonably.
I find downvote visual cues such as the graying out a bit like laugh tracks in sitcoms. On their own, some people will find the jokes funny, some find them distasteful or unfunny -- add a laugh track and you're telling people how they must feel. Downvotes are a lot like that: "people find this comment distasteful, what are you going to do about that?".
But if they had said something like, "I suspect this post is written by GPT-3. Consider X, Y, Z sentence -- these are typical constructions of GPT-3 because of A," it probably would not have been down-voted.
But until spam becomes a problem on this site, it is premature to alter norms to protect against a non-issue.
Honestly, that's what keeps the quality of the site high. Compare that to reddit, where I've seen multiple responses that were literally just "bruh"
Edit: I should say mostly reasonable. Person 2 ended with uncalled-for snark, and as I said, person 1 potentially was being unnecessarily sarcastic. But I don't think this interaction is a good example of "bullshit".
> I had a lot of fun doing this experiment. Some people will probably get mad at me for tricking people, but whatever.
OP even recognizes the externalized cost of their actions, "but whatever", as they say. What they did by posting fake content to prove a point was wrong, betrayed everyone's trust, and was completely against the spirit here. And as a community we could say that we don't support this behavior. I think that's a good idea.
Person2 was very dismissive without even addressing the comment. That to me seems antithetical to "trying to uphold the tradition of trust and charity," though I'm pretty new here. I personally don't think a dismissive attitude and a snarky quip for no good reason is worth anything.
> YIKES.
This.
These condescending comments is what makes the community aspect of any kind of website terrible. I've felt for a long time that using the net without joining a community and leaving comments is the healthiest way to go about it. I'm not trying to take up the moral highground, I'm guilty of behaving as a douche, but ever since I started refraining myself from commenting - my internet experience and mental health has been way better and even the douche aspect started to tone down.
Knowledge about this has to be hidden from everyone, though and I don't know how long it can be kept hidden.
I'd actually like to see HN try this for a day. Some of the most substantive content is also the most voted down content.
Meh, HN already has hidden functionality that they don't tell users about, and honestly it feels like psychological torture. For example, all of my posts lose points after posting. For a long time, I thought a bot was following me and downvoting me. No, I was just being tortured by dang, feeling like I was crazy for all of my posts receiving down votes. Unless I consistently only post things that subscribe to the norm, I will be in a points deficit. I will never be able to downvote posts and participate in the community at large. I've effectively been cancelled, mostly for having different opinions and partly for responding to trolls in an unhelpful way. I'm working on the second part, but I don't think that's the problem.
I'm not interested in trying to psychologically harm people, especially in a time when social activities only really exist on the internet.
How else are you supposed to respond to nonsense other than saying "This is nonsense"?
It's also why adequate disclosure of AI-generated text is important and it's good OpenAI is advocating that.
I agree that it's an edge case. I don't think moderators would have replied to https://news.ycombinator.com/item?id=23894742 as a guidelines violation, but the user who did meant well and all this stuff is pattern-matching anyhow. For example "this sounds like GPT-3" can just as easily be an internet insult (e.g. "sounds like GPT-3 regurgitated a sociology textbook" - https://news.ycombinator.com/item?id=23687199), in the same way that "this sounds like a Markov chain" used to be. The fact that that comment added "or the human equivalent" kind of made it lean that way. Perhaps that's why it provoked an immune response from a community member while the more neutral "I think this was written by GPT-3" comment did not.
It's your choice if you want to invest that time and effort, of course. But you can also choose not to leave low-effort comments. Brandolini's law comes to mind.
This got me wondering. Just like many GAN-generated human faces are actually very similar to actual faces in the training set, is it possible that this blogpost is quite similar to something someone actually wrote and was part of the GPT-3 training set?
[1]https://adolos.substack.com/p/feeling-unproductive-maybe-you...
The biggest benefit of GPT-3 is efficiency. All I need to write is a good title and intro. I could write five of them in an hour and publish them all in one day if I wanted to.
The author admits that fact checked stuff can't be written this way. So you can't really produce solid writing this way. Yet it is being proposed as the future of content creation.
As someone who does some content creation, I really dislike this idea that this is "efficient" and "can save money." Good writing is increasingly hard to find as is, in part because we don't pay writers what we used to pay them. Good writing involves research and understanding the topic. It's not just about putting pretty words together.
I don't think this is a good trend.
In my case, I have plenty of stories I want to write, lots of ideas that I want to play with. The thing I struggle with is writing the first draft of a story - it turns out that I find writing the first draft of a story to be really, really boring work (even with the help of NaNoWriMo). My fingers are typing out the words for chapter 3 but my brain is already grappling with the obvious plot holes in chapter 28.
This is where GPT-3 could really help: churning out words to fill the page with crap first draft copy. Feed the existing copy into the GPT-3 engine, alongside some prompts about the directions I want the story to go, themes I want it to tackle, etc. Receive back 2-3 drafts of the next thousand-15 hundred words. Then edit the heck out of those suggested words to bring the crap copy into line with my vision for the chapter.
I like this idea because most of my storytelling work happens in the first and second revisions, not the initial draft. I'd be happy to call the results my work, not the work of some mindless ML bot. Proper writers will probably feel differently about the process.
In my scenario, I expect there will be a lot of books written by ML algorithms, then published/sold to readers with inadequate (or no) additional author/editor input. My bet for the first fiction genre to take this approach is Romance Pulp - particularly Mills&Boon-style novels which have been written in line with a tight template for decades. And ML 'authors' in the literary fiction genre - it wouldn't surprise me if these already exist!
(Shrug)
I suspected GPT-3, but the lingering throught that it was written by a real person made me back down.
Well, at least I feel a little validated. And also concerned about the absolutely nonsense that gets promoted or upvoted by people that either don't read the articles, or just don't pay attention.
Edit: I recognise what you are complaining about but that low quality is not entirely the fault of the people who leave those low quality comments. They are encouraged to leave poorly researched comments on subjects they don't understand very well exactly because the average quality of comments on the subject of machine learning and AI is so low. People can see that almost everyone else is making it up as they go along or just trying to show off knowledge they don't have, so they think, "that's the done thing, I should do the same". If more knowledgeable users commented more often, the overall quality of the discussions would improve and less experienced users would be encouraged to think more about the quality of their own comments. And of course, if more knowledgeable uses commented more often, the average user would also learn more about machine learning and AI than they are likely to learn from the articles that are typically posted on HN about those subjects (and the discussions of them). Everyone would benefit from that, not least knowledgeable users who would have an opportunity to discuss these subjects at a higher level than they can now.
Generally my opinion is: if you have specialised knowledge that people are interested in, spread it around. Only good things can come from that.
About a year ago some international business trainer made a splash in Moscow, Russia, selling out the whole stadium of quite expensive tickets. His show was an utter bullshit, people made memes with his quotes like "Do right things, and don't do wrong things" or "Ask with question, but answer with answers" and so on. But people who bought tickets to the show were still satisfied, because people in need of business trainer's advice are obviously in a very vulnerable position and will take any shoulder pat as a sign of support.
Next time that life coach should put a GPT-3 on stage.
I think the author is not making a clear enough distinction between writing and generating "content". Content has become a hollow commodity in the new media landscape. It's purpose it to generate clicks, not inspire on a deeper level. GPT-3 seems much more geared towards the content side vs the expressive side that defines great written pieces.
Writing is so much more about the mental process of refining ones thoughts, rather than the act of scribbling lines onto a page. Writing is hard because it forces clarity in order to be effective. I don't see how using an AI tool to outsource the heavy lifting of reflection and rumination can be considered a true "writing" tool.
In addition, it's also a big deal to me how rational and objective commenters on this site are in comparison to most other websites these days. As a hardcore free-speech fan and proponent of objectivity, it's a nice a breath of fresh air.
I've found that communities tend to develop a sense of self-importance and overrate their own "good" qualities. I remember when I used to read Slashdot, more than a decade ago, many users would write things like "of course, all of us who read Slashdot are above average users and $PROBLEM does not affect us, but consider a typical Joe User". When I read facebook groups about niche topics I find interesting, inevitably a member will claim "don't you find this is the best facebook group about $TOPIC? We don't argue, we have nice reasonable conversations about $TOPIC, and the mods are top-notch!" (similar groups about the same topic will of course each claim to be the most reasonable, well-moderated, argument-free group). And the list goes on.
Unsurprisingly, many HN regulars tend to hold the same views about HN ("the most rational, less biased, most diverse ideas"), possibly unaware that most communities tend to develop the same feelings about themselves. Human nature, I guess :)
Do note in practice I've settled on HN for tech news, for good or bad. My time is finite, I cannot check a thousand sites :)
I'm a newbie, but the reason I came here was for more signal than noise and several times in the last months I've considered looking for new places because the comment quality is slipping.
None of it is discussing the content, which makes sense given that it was nonsense.
It just makes me further believe a very small minority read the article before jumping into discussion. Like that other GPT-3 post where a ton of the commenters didn't make it to the punchline in the article.
I guess that points to the value of a clickbait headline too.
I remember skimming the article and thinking it was nonsense, but downvoting the "this is GPT-3 nonsense, or the human equivalent," comment. I don't see a contradiction there.
Last, I totally disagree with the author's premise quite strongly. Maybe if you're trying to just pump out clickbait where the whole point is the headline, it'll make a difference in your ability to churn out content.
> However, with very minor editing, you could make these decent. Cut out irrelevant stuff, write a conclusion, and boom - people don't stand a chance of telling the difference.
You can give it a nice form, sure. People might believe it, too, but it'll still be crap.
> The difference is, this is one of the first times technology has the potential to affect jobs in the creative sector.
This is really wrong. MS word, spell check, printers, photoshop, grammar check, the ipad, digital cameras, practically everything in movies and animation these days. Technology has been helping creatives be more efficient for ages. One person with photoshop can do work in a day that would have taken ages not long ago. You can hire Eddie Murphy once instead of a full cast. You can film the mandalorian with massive LED monitor backgrounds instead of a greenscreen, which itself replaced other work.
I guess if your goal is to churn out thought pieces with a nice headline, but zero research and little critical thought, it's going to change the game. All you need to do is say whatever you need to fit your narrative, like the fact that computer technology and automation is just finally starting to affect creative jobs. It'll flow perfectly and people won't be able to tell it's all made up. Maybe the author's right and it's the sad reality that this is what most of the content we read is and what most of the internet content writing jobs consist of.
I did think it was a human. I just thought it was low quality selfhelp stuff that you see out there. The "...too productive" one was more convincing because it felt like more of a personal story that didn't need to have references.
(punting on how "expertise" is measured or assigned)
HN does seem to fetish certain orthodoxies.
I am actually proud that the community stood up for the ideals of trusting others (even if the trust was misplaced). I am also pleasantly surprised at how well GPT can mass produce such articles.
I'm not sure this is a bad thing. I tend to read just the comments because they're ultimately more interesting. Blogs/articles are all fighting for attention in the world and they tend to be bombastic or just so explanations or other simplistic interpretations. HN comments tend to have the "real story". You get anecdotes and different perspectives and more nuanced opinions.