Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google is picking ChatGPT responses from Quora as correct answer

twitter.com

Post reply on HN
Print view View original

81–90 of 239 posts

Re: Google is picking ChatGPT responses from Quora as correct answer

#81

Post by bagasme »

jsnell wrote: post #23

What I don't understand is why Quora would be auto-generating an answer for this question in the first place. The theory in the tweet that it's about identifying a frequent query doesn't really explain anything: the question already had two answers written by humans years ago, there was no need for them to come up with additional answers.

And when Quora also shows answers from "related" (but IMO not strictly related to the original question in question) by default, I think these answers are mostly incorrect when being applied to the original question.

Re: Google is picking ChatGPT responses from Quora as correct answer

#82

Post by LetsGetTechnicl »

giancarlostoro wrote: post #52
rob74 wrote: post #13

Google's "answers" feature was already unreliable enough when there was only one AI involved (the one picking the answers) - I remember it once said that the capacity of an Amiga floppy was 1760 KB, because the Wikipedia page it picked ( https://en.wikipedia.org/wiki/Floppy_disk_variants#Amiga ) failed to mention the capacity of the "standard" floppies and only provided a number for the (non-standard) HD floppies.

I once googled a niche question, liked the instant answer, saw it was on HN. Clicked on it. It was me. I wrote the answer to my own question, and I guess forgot 5 years later.

That reminds me of Twitter user @foone who often goes on deep dives on random tech and old chips, and on several occasions the top Google result for a chip serial number is one of their old tweets

Re: Google is picking ChatGPT responses from Quora as correct answer

#83

Post by crazygringo »

It seems like the obvious solution here is to voluntarily label AI content (text, image, anything) with some kind of HTML wrapper or attribute that search engines can then choose to index differently.

Obviously lazy/deceptive people can choose not to label it, but in a case like this, Quora is already visually labeling the answer as AI-generated for users, so it seems like they'd be willing to tag it in HTML as well.

Have there been any W3 proposals on this? Any potential standards?

Re: Google is picking ChatGPT responses from Quora as correct answer

#84

Post by jacurtis »

jsnell wrote: post #23

What I don't understand is why Quora would be auto-generating an answer for this question in the first place. The theory in the tweet that it's about identifying a frequent query doesn't really explain anything: the question already had two answers written by humans years ago, there was no need for them to come up with additional answers.

Quora is already a dumpster fire. Everytime I go to quora from Google, I see the question at the top, but then right below it I see answers for different questions. You have to scroll past those, past a block of ads, past a sponsored answer (to another unrelated question), just to get to the answer, which itself is an confidently incorrect human that is just peddling their own website for SEO juice, and it is truncated at 2 lines, which means you need to click "read more" which spawns an additional set of ads that you need to scroll past to read lines 3 and 4 of the question.

I visited a link this week and saw ChatGPT as the top result and I thought the same thing. Is Quora signaling to me that you are better off not asking our community for questions, just throw it to the AI gods and good luck. Oh and look at these 3,000 ads and take these 209 cookies before you go.

Luckily I search with Kagi now, so I told it to just never give me search results with Quora in it. I'd prefer not knowing that dumpster fire of a website still exists.

Re: Google is picking ChatGPT responses from Quora as correct answer

#85

Post by jondwillis »

giancarlostoro wrote: post #52
rob74 wrote: post #13

Google's "answers" feature was already unreliable enough when there was only one AI involved (the one picking the answers) - I remember it once said that the capacity of an Amiga floppy was 1760 KB, because the Wikipedia page it picked ( https://en.wikipedia.org/wiki/Floppy_disk_variants#Amiga ) failed to mention the capacity of the "standard" floppies and only provided a number for the (non-standard) HD floppies.

I once googled a niche question, liked the instant answer, saw it was on HN. Clicked on it. It was me. I wrote the answer to my own question, and I guess forgot 5 years later.

Has happened to me on StackOverflow multiple times. Always hilarious and makes one reflect.

Re: Google is picking ChatGPT responses from Quora as correct answer

#87

Post by sanj »

giancarlostoro wrote: post #52
rob74 wrote: post #13

Google's "answers" feature was already unreliable enough when there was only one AI involved (the one picking the answers) - I remember it once said that the capacity of an Amiga floppy was 1760 KB, because the Wikipedia page it picked ( https://en.wikipedia.org/wiki/Floppy_disk_variants#Amiga ) failed to mention the capacity of the "standard" floppies and only provided a number for the (non-standard) HD floppies.

I once googled a niche question, liked the instant answer, saw it was on HN. Clicked on it. It was me. I wrote the answer to my own question, and I guess forgot 5 years later.

I had a similar experience trying to find a justification for a “House rule” we use when playing Catan (aka “the penny rule”).

In trying to find a reputable source, the first result was to an HN comment… that I wrote.

The folks we were playing with were not convinced.

Re: Google is picking ChatGPT responses from Quora as correct answer

#88

Post by aftbit »

crazygringo wrote: post #83

It seems like the obvious solution here is to voluntarily label AI content (text, image, anything) with some kind of HTML wrapper or attribute that search engines can then choose to index differently. Obviously lazy/deceptive people can choose not to label it, but in a case like this, Quora is already visually labeling the answer as AI-generated for users, so it seems like they'd be willing to tag it in HTML as well.…

It would be lovely if this were required for advertising as well. That would make adblock so much easier. Maybe there will be aiblock in the future?

Re: Google is picking ChatGPT responses from Quora as correct answer

#89

Post by 8teapi »

capableweb wrote: post #56
8teapi wrote: post #53

Earlier quoted context omitted.

They’re using text-davinci-003 which is a static instruct endpoint. The 3.5/4 ChatGPT endpoints are dynamic and continuously updated. You magically find buggy behavior fixed when using those.

> They’re using text-davinci-003 which is a static instruct endpoint. Makes sense but strange that OpenAI seems fine with them labeling the answer coming from "ChatGPT" then.

OpenAI has been fairly loose on the branding.. They never expected ChatGPT to be huge and a bunch of devs were already using the davinci APIs. Earlier on the ChatGPT API was not even available for external use. Plus ChatGPT became famous but not OpenAI so much. So they told devs you can add powered by ChatGPT if you use the API endpoints.

Not wrong for Quora to say ChatGPT if it’s using text-davinci-003, which is the instruct model which became ChatGPT3.5.

Re: Google is picking ChatGPT responses from Quora as correct answer

#90

Post by goodluckchuck »

userbinator wrote: post #73

This happened not long ago: https://news.ycombinator.com/item?id=37368243 We were already in the "post-truth" world, and AI is just going to drive us further down that path. The tendency of a lot of the population to prioritise quick and vapid answers over researching and thinking for themselves isn't helping either. That said, I'm not sure why whether eggs can be melted is a popular question. Is that a meme?

Post-truth is the greatest step toward truth in human history. For centuries the clergy, government, media, and corporations have owned the Truth and defined the Overton window. Post-truth’s overwhelming flood of true, false, and irrelevant claims is empowering critical thinking. It’s offering freedom.

The printing press led to the renaissance. The internet is now leading a new one.

There will continue to be protests by the establishment powers, and unfortunately there will likely be conflict. The fall of monarchies was not without bloodshed. Hopefully we can find a way to achieve similar progress this time around without the same loss of life.

Post reply on HN