College student uses GPT-3 to write fake blogs and ends up top of Hacker News | Hacker News Reader

College student uses GPT-3 to write fake blogs and ends up top of Hacker News

theverge.com

40 pointsfactordaily_22 comments

https://news.ycombinator.com/item?id=24165040

... wherein dang points out that "... the story is bogus and based on false claims."

Added in edit after the first reply to this ...

Also quoting dang:

"They've been trying to manipulate HN with multiple accounts and voting rings—which hasn't worked—and baity titles, which unfortunately has."

Those seem like small technicalities. GPT-3 needs a prompt to write something so a few sentences have to be written by a human. Also, whether 1 or 2 people pointed out that it could be GPT-3 is not a substantative difference. The screenshot shows a greyed out comment which means it was downvoted. Even assuming the screenshot is faked, those comments are still toward the bottom of the thread so weren't ones that many people agreed with. Overall, HN was gamed and I hope this will make people think more about the quality of content they upvote here. [This comment was generated by GPT-3] [just kidding]

Link to the article: https://news.ycombinator.com/item?id=23893817

https://news.ycombinator.com/item?id=23894742 And the call that it was gpt 3 with the complaint about this being rude

Haha! That's hilarious

Plot twist: jackkinsella is actually a GPT-3 run account.

Plot twist: I'm actually a GPT-3 run account

How posts on HN reach #1 is a topic all by itself. To assume it's impossible to game the ranking is foolish. If I were going to try to fool people with an AI generated story, I might also take special steps to help push that post as high up as possible.

Regarding the "content" of the post in question, as TheVerge article notes, "there’s a lot of not-great writing on these here internets, so I guess it’s possible that this could pass as “content marketing” or some other content".

Indeed; so much of the internet is full of meaningless fluff and re-speak (for SEO gaming, referral link promotion, etc.), that whether the fluff is generated by AI or a human is kind of irrelevant.

Funny how someone pointed out that it lacked content and seemed like something written by GPT-3 and got shit for it

ahaha classic hacker news, missing the point elevated as an art form.

I nearly always read the comments first and in most cases don't read the article. I don't up-vote often, but I'll occasionally up-vote an item which appears to be developing an interesting discussion.

There's also a lot of cases where the article could be an arbitrary choice for an explosive news item. I view HN as more of a forum by news topics than an aggregator of news links.

An article which suggests that HN gets fooled by AI doesn't know how HN works. Though most media articles seem to misunderstand technical topics in general.

Good point. The title and topic of an article if interesting generates discussion and input. Does not necessarily have to do with any of the article content.

Lots of people don't even bother reading DFA so you don't even need gpt-3 to "fool" HN.

I suspect a large part of what gets people to the front page are the first 10-20 votes and a large portion of those probably come from people who only read the title (because the first 10 come fairly quickly).

But I also realise I don't have the data to back it up. Does anyone know of a dataset that has upvote times in it?

What does DFA stand for?

Da Fine Article

I see that ducking autocorrect hates you, too.

How is this not going to turn the internet into a junk box?

This is my concern as well. Do we need to create the opposite of this? A service that searches the internet and 'wades' thru information/trash and then reports back with 3 best possible conclusions?

Let's hope Google has the antidote? Otherwise they are doomed too

I think too much of my HN consumption and upvotes is based on title. I know what will be in the article when I just read "Feeling unproductive? Maybe you should stop overthinking". If I agree with the thesis, upvote based sites like hacker news encourage quick upvotes.

I think A.I. chatbots have been in online discussion groups for decades, if not the beginning, e.g. Eliza. They are posts that almost make sense, but are off the mark again and again. A current candidate is Craiglist forum astrophysics frequent poster NumberlineA.

Reply on news.ycombinator.com