I, ChatGPT
-
@Arantor said in I, ChatGPT:
But I also think we'll see a return to the human-curated directories of old as the next major iteration of discovery for the simple reason that the arms race is, as far as I can tell, about to be lost to the machines.
This, except that the curating will be done by AI.
And no, I'm not joking for the fun of a self-referential joke.
This might be a case where AI can actually work reasonably well, because the goal is not to give one authoritative and correct answer, but to fuzzily filter the metaphorical wheat from the chaff. It doesn't matter if a bit of chaff remains, or some wheat is lost, as long as what you get is more wheat than chaff (we're talking search results here, where you are going to pick what you want from a set of results afterwards).
AI content will manage to pass through but if it does and does so by being actual interesting content (or content that answers your query), then why would that be an issue?
(honestly, I'd be surprised if Google was not already using, or at least testing, AI-as-a-search-engine -- and I'm not just talking about wrapping the traditional search result in a nice sentence written by AI but really using AI to generate the search results)
At that point, I'm not sure what the next step in the arms race will be. AI content farms that actually become more and more correct/useful to keep being selected? That would actually be a net positive, so that can't be what will happen. So I don't know.
-
-
@dcon Already sorted.
-
@remi said in I, ChatGPT:
honestly, I'd be surprised if Google was not already using, or at least testing, AI-as-a-search-engine
Last I heard the Google one wasn't really working, but Microsoft thinks their works well enough to
. It is TANGO ROMEO ALPHA SIERRA HOTEL.
-
@Bulb the ones I heard about looked more like a layer of AI that wraps over search results rather than powering the actual search itself.
But then again, since nobody knows how the search itself actually works, maybe their attempt at "AI search" is actually doing that.
-
@remi From what the Bing one produces, it will take your question, produce (and show) the terms to actually hand over to the search engine, then tries to distill answer from the results, referencing them like citations.
In the few tries I gave it, it tended to just write the answer I considered likely and referenced the obvious documentation for it.
Except, me being me, I tried it after I already read that documentation and noted that it does not actually say the thing I wanted to know, so the results were rather useless to me. Worse then useless, because what it confidently claimed the documentation says not only the documentation didn't say, but it also wasn't actually true.
… for the quick search for common things just so you don't have to remember how that thing was it probably helps a bit, but for advanced things it does not … so when the junior monkeys fully upgrade from copy&pasting from stack overflow to copy&pasting from bing chat, the carnage is going to be gory.
-
@remi said in I, ChatGPT:
actual interesting content (or content that answers your query)
These two are not equivalent. Anyone can produce content that answers your query but is not correct. See any amount of content from the ongoing disinformation wars.
-
@Arantor said in I, ChatGPT:
@remi said in I, ChatGPT:
actual interesting content (or content that answers your query)
These two are not equivalent. Anyone can produce content that answers your query but is not correct. See any amount of content from the ongoing disinformation wars.
You could have just pointed at the flamewars here over trival shit to be honest.
Remember folks. Our policy concerning the oxford comma is that you will be laughed at for using it.
-
@DogsB said in I, ChatGPT:
Remember folks. Our policy concerning the oxford comma is that you will be laughed at for using it.
But also for not using it, and any other reason we feel like.
-
@Arantor said in I, ChatGPT:
Little known fact, there was a rival to Wikipedia back in the day, the Open Encyclopedia, or Open Site as it was known, and was essentially operated under the same thinking as Dmoz, that it was hierarchical and people got editing rights for tiny leaf branches and earned more over time, with more senior editors doing crazy things like discussing if the changes were sensible.
I do often wonder what the alternate future looks like where Wikipedia didn’t take off and OE did.
At least h2g2 is still around:
h2g2 - The Guide to Life, The Universe and Everything
h2g2, the Unconventional Guide to Life, the Universe and Everything
-
@loopback0 I have acquired the habit of using it since I started participating here.
-
@remi said in I, ChatGPT:
At that point, I'm not sure what the next step in the arms race will be. AI content farms that actually become more and more correct/useful to keep being selected? That would actually be a net positive, so that can't be what will happen. So I don't know.
The immediate "war" will be between exactly what you allude to. AI system generating stuff vs AI system filtering stuff. Making content "useful" isn't the only solution for the former, though. The latter won't be perfect, so there will be ways to smuggle "useless" content past it. In that sense, it's the same story as with SEO and so on. "Optimizing SEO" (or whatever they call it) was never synonymous with providing better content.
It's also essentially what people do with adversarial networks all day.
-
With a bit of chance, AIs will be so busy fighting each other that they'll spare us humans.
-
@Arantor and ChatGPT did a shit load of a better job on this than Fall Out Boy did…
-
@cvi I fully expect the generating AI to use the adversarial content generation, just like you have adversarial images—the kind specially prepared so that every human sees an apple, but the specific image recognition AI will keep saying it's a banana. So humans will still see ad-ridden clickbaits.
-
@remi said in I, ChatGPT:
At that point, I'm not sure what the next step in the arms race will be. AI content farms that actually become more and more correct/useful to keep being selected?
Cueue XKCD #whatever.
-
@Bulb said in I, ChatGPT:
So humans will still see ad-ridden clickbaits.
The problem is that you don't need any special trick to end up with this. Search engines being ad-driven businesses, their metric for "our AI is good at selecting results" will be "what generates the most clicks."
Which in an ideal world would be "good content" but in the real world is "content that users like seeing" and that's a subtle but important difference, because this is how we end up with clickbait (click bait works because people voluntarily decide to click on them i.e. it's something they want to see).
In an economically ideal world (or at least a perfect market economy, whether that is an ideal world is best discussed elsewhere...), this could possibly be solved in the long term by new search engines that use another metric (possibly because they'd use another business model, you could imagine a non-free search engine for example) appearing and taking over the market. In the real world though, the network effect and bundling of search with other products and monopolistic practices etc. all mean it's unlikely to happen.
So we're doomed but everybody will keep on living regardless.
-
@Arantor said in I, ChatGPT:
@Arantor and ChatGPT did a shit load of a better job on this than Fall Out Boy did…
Fall Out Boy were a crime against good taste in music.
On topic but not related to the quoted post... There is actually a huge market for generic comfort food. Ye have seen my reading list. There’s a couple series of books there that could be hammered out by Chatgpt that I would happily hoover up.
Mills and Boons have had thirty authors shitting out a variation of the same four stories for about 50 years and they sell by the bucketload. Don’t think anyone would notice a move to ChatGPT there to be honest. There must thousands of photos of Fabio to train an AI on to create the book covers.
-
@Zecc said in I, ChatGPT:
@remi said in I, ChatGPT:
At that point, I'm not sure what the next step in the arms race will be. AI content farms that actually become more and more correct/useful to keep being selected?
Cueue XKCD #whatever.
810
-
@DogsB said in I, ChatGPT:
There must thousands of photos of Fabio to train an AI on to create the book covers.
-
@DogsB said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@Arantor and ChatGPT did a shit load of a better job on this than Fall Out Boy did…
Fall Out Boy were a crime against good taste in music.
-
@remi said in I, ChatGPT:
"what generates the most clicks."
I'm not in the search engine business, but wouldn't there be a lot more other factors? I.e., it has to be clicks to the "right" content (i.e., one that they are being paid for). This doesn't necessarily align with the people generating content (unless they also pay the search engines, but they'd rather not pay for their clicks, so there's that).
That said, you're right in the sense that all parties involved have a stake in screwing over the users. It's more that these don't necessarily align with each other.
Filed under: Isn't it great, being the product?
-
@cvi you're right that the number of clicks by itself is just one part of the picture, but it's a simple shortcut that conveys the big idea, i.e. that what is "good" for the search engine is not necessarily what is "good" for the user of the search engine.
That said, you're right in the sense that all parties involved have a stake in screwing over the users. It's more that these don't necessarily align with each other.
It's not that they have a stake in screwing the user, it's just that their goals don't align with the user's goal. But that said, we agree with each other on the whole.
Note that this isn't really new, it's the case for most companies. Their goal is always to sell you something, whereas your goal is to fulfil whatever need their product fulfils. Usually when you're directly paying for something that something has to be pretty closely aligned with your goals, so the gap usually isn't too big. But this isn't the case here, the alignment only has to hold enough for you to not stop using the search engine, which is a pretty low bar and easily influenced with e.g. network effect, bundling with other tools and similar things.
Filed under: Isn't it great, being the product?
But it's free!11!
-
-
@ixvedeusi said in I, ChatGPT:
@remi said in I, ChatGPT:
But it's free!11!
Filed under: Nothing ain't worth nothing
Freedom isn’t free,
It costs folks like you an’ me
And if we don’t chip in our buck o’ five, who will?
-
@remi said in I, ChatGPT:
@Bulb said in I, ChatGPT:
So humans will still see ad-ridden clickbaits.
The problem is that you don't need any special trick to end up with this. Search engines being ad-driven businesses, their metric for "our AI is good at selecting results" will be "what generates the most clicks."
Which in an ideal world would be "good content" but in the real world is "content that users like seeing" and that's a subtle but important difference, because this is how we end up with clickbait (click bait works because people voluntarily decide to click on them i.e. it's something they want to see).
They voluntarily decide to click on them, then realize their mistake, because it was a click bait that looked, from the title, like it might say something they want to know, but actually does not, but the search engine has already recorded the click, so it will continue to provide that kind of content even though the users would actually prefer if it didn't.
-
@Bulb said in I, ChatGPT:
@remi said in I, ChatGPT:
@Bulb said in I, ChatGPT:
So humans will still see ad-ridden clickbaits.
The problem is that you don't need any special trick to end up with this. Search engines being ad-driven businesses, their metric for "our AI is good at selecting results" will be "what generates the most clicks."
Which in an ideal world would be "good content" but in the real world is "content that users like seeing" and that's a subtle but important difference, because this is how we end up with clickbait (click bait works because people voluntarily decide to click on them i.e. it's something they want to see).
They voluntarily decide to click on them, then realize their mistake, because it was a click bait that looked, from the title, like it might say something they want to know, but actually does not, but the search engine has already recorded the click, so it will continue to provide that kind of content even though the users would actually prefer if it didn't.
All according to ai-kaku
-
-
ChatGPT owner hit by class action lawsuit over alleged stolen data
Digital chatbot trained by ‘secret scraping of the internet’ violating privacy laws, accusers claim
The lawyers are coming!
-
@DogsB said in I, ChatGPT:
ChatGPT owner hit by class action lawsuit over alleged stolen data
Digital chatbot trained by ‘secret scraping of the internet’ violating privacy laws, accusers claim
The lawyers are coming!
Lawyers want to get in on the gravy train too
-
@izzion and the lawyers have been fined making a precedent that if you use AI to do your research, you better check it’s accurate before using it in court.
-
https://forums.theregister.com/forum/all/2023/07/01/microsoft_github_copilot/
Github is about to see an update to their TOS.
*edit actual article.
Microsoft and GitHub try to derail Copilot code lawsuit
And so far, they might succeed: Where's the smoking gun?
-
@DogsB said in I, ChatGPT:
https://forums.theregister.com/forum/all/2023/07/01/microsoft_github_copilot/
Github is about to see an update to their TOS.
That’s an amazing defense: “we copied everyone’s code, so you can’t point to anything in particular we copied.”
The companies maintain that because GitHub users decide whether to make their code public and agree to terms of service that permit the viewing, usage, indexing, and analysis of public code, then the site's owners are within their rights to incorporate the work of others and profit from it.
"Any GitHub user," they say, "... appreciates that code placed in a public repository is genuinely public. Anyone is free to examine, learn from, and understand that code, as well as repurpose it in various ways. And, consistent with this open source ethic, neither GitHub’s TOS nor any of the common open source licenses prohibit either humans or computers from reading and learning from publicly available code."
That is absolutely not how copyright works!
That’s almost a
argument, “I only read the code, not the license terms.” You can’t just do whatever you want with it unless the license says so.
-
Many GitHub repos do indicate their licence state in a machine readable fashion so odds are that’ll just be used aggressively for slurping.
-
@Arantor said in I, ChatGPT:
Many GitHub repos do indicate their licence state in a machine readable fashion so odds are that’ll just be used aggressively for slurping.
And their response is akin to saying “we don’t care if it’s GPL or MIT or whatever, we’ll treat it as public domain.”
-
@topspin said in I, ChatGPT:
That is absolutely not how copyright works!
That’s almost a argument, “I only read the code, not the license terms.” You can’t just do whatever you want with it unless the license says so.The problem that all F/OSS faces, especially the FS part of F/OSS, is that lots of people, including some who ought to know better(1), think that "GPL licenced" is equivalent to "in the public domain".
Hint for those who read this in the fullness of their ignorance: No. The core principles of the GPL, and Free (as in liberty) Software in general, rely on the existence of copyright (even if they call it "copyleft") in order to be able to apply the restrictions that they apply to their liberty-mode-free software.(2)
Yes, folks, "'free software" in the GPL sense is a contradiction in terms. I worked this out within seconds of reading the terms of the GPL (v2) and RMS's "GNU Manifesto" in 1995 just after my first time installing a flavour of Linux-based operating system.
(1) Including someone who posted on the pre-Americanisation version of The Register, trying to incite people to "abandon copyright and embrace the free software". I reamed him good and proper, although without mentioning the GAU-8 in my back pocket.
(2) Yes, there's a teeny bit of sarcasm there...
-
@Steve_The_Cynic said in I, ChatGPT:
Yes, folks, "'free software" in the GPL sense is a contradiction in terms.
It's a vampire. If you're writing commercial software, you should not touch GPL-licensed software (and other licenses, but GPL is probably the worst) without prior written approval from your legal department.
-
@PleegWat I tend to refer to it as a virus since GPL has a habit of trying to make everything it touches follow the same restrictions. People don't believe me when I tell them this.
-
-
@HardwareGeek said in I, ChatGPT:
Hello, I'd like to introduce you to AGPL.
: Sorry, I'm already a believer of another religion.*closes the door*
-
@HardwareGeek it makes sense though because the use case Affero “solves” is the one use case GPL didn’t really ever account for, namely that this “web” thing would exist and take off (especially since GPLv2 dates from June 1991 and Berners-Lee’s hypertext proposal didn’t get made publicly visible outside CERN until August), and up to that point, there were tools for things like BBSes and FTP etc that were free in some fashion.
Just as Tivoisation birthed GPLv3, it was inevitable that web server style applications would get targeted eventually.
-
In today’s episode of that’s not how copyright works, you know it and we know it:
Google Says It'll Scrape Everything You Post Online for AI
An update to Google's privacy policy suggests that the entire public internet is fair game for it's AI projects. If Google can read your words, assume they belong to the company now, and expect that they’re nesting somewhere in the bowels of a chatbot.
In unrelated news: I need to find a different email provider.
-
@topspin Since when has Google ever cared about copyright? Have we learned nothing from YouTube or Google Books?
-
@topspin said in I, ChatGPT:
find a different email provider
May suggest Yahoo! ?
Waht about Hotmail?
Or would you prefer AOL? Since you do not live in Scunthorpe, you may be able to register.
Then there are gmx.de, web.de, freenet.de, t-online.de, ...
-
-
The careers that will survive the AI revolution – and pay the highest salary
Manual labour jobs and roles requiring subjective reasoning are deemed the safest from automation
Prompt engineers, AI ethics officers, and AI auditors are expected to be common job roles in the near future, Adzuna said.
After the initial grift has past, prompt engineering is going to be labeled a soft skill like googling.
Ethics will be a good racket if you can get in on it. We have those for social media. Their requests are never listened too but they are trotted out for puff peices every other week.
Noone knows how it works so Auditing is going to be another grift like SEO.
-
@remi said in I, ChatGPT:
AI content will manage to pass through but if it does and does so by being actual interesting content (or content that answers your query), then why would that be an issue?
For the same reason why urban legends, bad science and fake news passing through is an issue, no matter how interesting they are.
-
7 advanced ChatGPT prompt-writing tips you need to know
Let's dive a bit deeper into the arcane art of AI prompt writing. There be secrets here, for those who dare enter.
Well thats some great prompt engineering.
take this two sentence autistic complaint about sunshine and expand it to a @remi style wall o text but make it sound as reasonable as @Arantor .
: 
-
@DogsB challenge accepted for when I’m not at work because trying to game ChatGPT is an interesting exercise and if humanity is already doomed, I’ll clap along just a little.
-
