I, ChatGPT
-
As ChatGPT gets “lazy,” people test “winter break hypothesis” as the cause
Unproven hypothesis seeks to explain ChatGPT's seemingly new reluctance to do hard work.
All hail the
!
-
@Applied-Mediocrity said in I, ChatGPT:
As ChatGPT gets “lazy,” people test “winter break hypothesis” as the cause
Unproven hypothesis seeks to explain ChatGPT's seemingly new reluctance to do hard work.
All hail the
!On Monday, a developer named Rob Lynch announced on X that he had tested GPT-4 Turbo through the API over the weekend and found shorter completions when the model is fed a December date (4,086 characters) than when fed a May date (4,298 characters). Lynch claimed the results were statistically significant. However, a reply from AI researcher Ian Arawjo said that he could not reproduce the results with statistical significance. (It's worth noting that reproducing results with LLM can be difficult because of random elements at play that vary outputs over time, so people sample a large number of responses.)
Meanwhile, OpenAI sucks up their moneys.
Speaking of...the CodyAI thing from Sourcegraph we've been using is powered by OpenAI. I'm a bit underwhelmed so far. I had it write me some unit tests. They were...OK. Basically tested stuff, though in a pretty different way than I would normally do it. Some of its bigger, block level autocomplete suggestions are OK-ish, too. Asking it questions about code seems to give decent answers. I suppose if I were less familiar with the code it might be more useful to me.
-
@boomzilla said in I, ChatGPT:
I suppose if I were less familiar with the code it might be more useful to me.
Or less useful, if it gave you wrong answers that you failed to detect were wrong due to your lack of familiarity.
-
Here comes the proctoscope!
Microsoft's OpenAI investment might merit a closer look
Hey Bing, how can I invest billions in a company but not break antitrust laws?
-
Amazing how they get offended when someone is stealing their content…
ByteDance is secretly using OpenAI’s tech to build a competitor
/cdn.vox-cdn.com/uploads/chorus_asset/file/25165917/Command_Line_Site_Post_ByteDance.jpg)
“They really just don’t want to get caught.”
-
@izzion it’s like ten thousand bots, …
-
-
...isn't it ironic, don'tcha think?
Well, no, not really.
-
Elon Musk’s Grok Twitter AI Is Actually ‘Woke,’ Hilarity Ensues
Grok has been launched as a benefit to Twitter’s (now X’s) expensive X Premium+ subscription tier, where those who are the most devoted to the site, and in turn, usual...
is taking immediate action 
Also, they mipelt Hillary
-
-
@loopback0 said in I, ChatGPT:
@hungrier said in The Official Funny Stuff Thread™:
So glad the AI takeover has made sales people and legal departments obsolete
-
-
-
-
This post is deleted!
-
Somebody wake up Pikachu...
GPT and other AI models can't analyze an SEC filing, researchers find
The findings from Patronus AI highlight some of the challenges of using AI models within big companies in regulated industries like finance.
-
@izzion said in I, ChatGPT:
Somebody wake up Pikachu...
The findings from Patronus AI highlight some of the challenges of using AI models within big companies in regulated industries like finance.
“One surprising thing was just how often models refused to answer,” said Qian. “The refusal rate is really high, even when the answer is within the context and a human would be able to answer it.”
I've had a lot of this. My stuff is apparently using ChatGPT 3.5 turbo, and often when I ask for it to generate some code it'll start, and then you get a comment like:
// put the rest of your code hereAnd it says something like, "This gives you an idea about how to do what you want..." So, yeah, super lazy.
-
-
@HardwareGeek AI is closer to human than we may think
-
I'm waiting for the day an AI will be able to
for me, so that I can spent my days ing instead.Huh, wait a minute...
-
@boomzilla said in I, ChatGPT:
I've had a lot of this. My stuff is apparently using ChatGPT 3.5 turbo, and often when I ask for it to generate some code it'll start, and then you get a comment like:
// put the rest of your code hereAnd it says something like, "This gives you an idea about how to do what you want..." So, yeah, super lazy.
It's internalised the Stack Overflow approach to giving answers.
-
@dkf said in I, ChatGPT:
It's internalised the Stack Overflow approach to giving answers.
Yeah, which is fine when I'm looking at SO and just need an idea about something I'm unfamiliar with. But the point of the AI in the IDE is to do stuff for me, especially the tedious stuff.
-
-
@dkf said in I, ChatGPT:
It's internalised the Stack Overflow approach to giving answers.
I thought the Stack Overflow approach was refusing to answer because of the question being a duplicate, or something like that.
-
@Zerosquare said in I, ChatGPT:
@dkf said in I, ChatGPT:
It's internalised the Stack Overflow approach to giving answers.
I thought the Stack Overflow approach was refusing to answer because of the question being a duplicate, or something like that.
Only if you don't have a good answer to suck up easy InternetPointzzz.
-
@Zerosquare said in I, ChatGPT:
@dkf said in I, ChatGPT:
It's internalised the Stack Overflow approach to giving answers.
I thought the Stack Overflow approach was refusing to answer because of the question being a duplicate, or something like that.
The ideal there (when there isn't a convenient duplicate) is to answer in a way that the asker can learn what to do, but without giving them a chunk code that they can just cut-n-paste without understanding.
-
@dkf I swore off SO years ago on the basis that any time I ever asked a question, the only person who'd ever answer it would be me, about a week later.
I later learned to use it as a really good benchmark for incompetence, c.f. another member of the dev team hits a blocker, 'I've posted it on Stack Overflow'... bad times ahead. (Mind you the person I'm primarily thinking of here got banned at least once for asking too many stupid questions)
-
@loopback0 said in I, ChatGPT:
@hungrier said in The Official Funny Stuff Thread™:
Bollox to that. You couldn’t pay me to take Chevy Tahoe.
-
@DogsB said in I, ChatGPT:
@loopback0 said in I, ChatGPT:
@hungrier said in The Official Funny Stuff Thread™:
Bollox to that. You couldn’t pay me to take Chevy Tahoe.
I'll take one!
-
@boomzilla said in I, ChatGPT:
@izzion said in I, ChatGPT:
Somebody wake up Pikachu...
The findings from Patronus AI highlight some of the challenges of using AI models within big companies in regulated industries like finance.
“One surprising thing was just how often models refused to answer,” said Qian. “The refusal rate is really high, even when the answer is within the context and a human would be able to answer it.”
I've had a lot of this. My stuff is apparently using ChatGPT 3.5 turbo, and often when I ask for it to generate some code it'll start, and then you get a comment like:
// put the rest of your code hereAnd it says something like, "This gives you an idea about how to do what you want..." So, yeah, super lazy.
when I tell it to give me the full code and not put any placeholder for brevity I usually get something more useful, but its with API + bettergpt.chat and gpt-4
-
Hooboy the 'Midjourney is plagiarising' folks are going to have a field day. MJ has always claimed that it doesn't keep the original art around for its systems therefore it can't be plagiarising but... v6 has some interesting behaviours.
https://twitter.com/Rahll/status/1737933582943871105
Attached to the tweet are several examples of 'the prompt is just the name of a painting' and what MJ spits out is a... fairly good reproduction of the painting, close enough to fool people at a first glance, anyway. It doesn't do the 'but it's not plagiarism' crowd any favours for sure.
-
@Arantor said in I, ChatGPT:
Hooboy the 'Midjourney is plagiarising' folks are going to have a field day. MJ has always claimed that it doesn't keep the original art around for its systems therefore it can't be plagiarising but... v6 has some interesting behaviours.
https://twitter.com/Rahll/status/1737933582943871105
Attached to the tweet are several examples of 'the prompt is just the name of a painting' and what MJ spits out is a... fairly good reproduction of the painting, close enough to fool people at a first glance, anyway. It doesn't do the 'but it's not plagiarism' crowd any favours for sure.
It would be interesting to see what can be spat out by something that has consumed all the art. On the other hand. Pay your fucking dues. The lawsuits should start kicking off next year so it should be interesting to see how it shakes out.
-
@DogsB the Twitter post I linked has some examples. In the interests of not making people go to Xitter I'll cross post.
There is an interesting argument being put forward that 'if I ask for a painting and I don't get something resembling the painting fairly closely, it's failed at giving me what I asked for' but there's no way to square that round peg without getting deep into the plagiarism stuff.
-
@Arantor said in I, ChatGPT:
Hooboy the 'Midjourney is plagiarising' folks are going to have a field day. MJ has always claimed that it doesn't keep the original art around for its systems therefore it can't be plagiarising but... v6 has some interesting behaviours.
https://twitter.com/Rahll/status/1737933582943871105
Attached to the tweet are several examples of 'the prompt is just the name of a painting' and what MJ spits out is a... fairly good reproduction of the painting, close enough to fool people at a first glance, anyway. It doesn't do the 'but it's not plagiarism' crowd any favours for sure.
There's a significant difference between deliberately plagiarizing something and claiming that because you looked at something and remember what it looks like everything you do is plagiarism. Those people can continue to get fucked as far as I can see.
-
@boomzilla said in I, ChatGPT:
@Arantor said in I, ChatGPT:
Hooboy the 'Midjourney is plagiarising' folks are going to have a field day. MJ has always claimed that it doesn't keep the original art around for its systems therefore it can't be plagiarising but... v6 has some interesting behaviours.
https://twitter.com/Rahll/status/1737933582943871105
Attached to the tweet are several examples of 'the prompt is just the name of a painting' and what MJ spits out is a... fairly good reproduction of the painting, close enough to fool people at a first glance, anyway. It doesn't do the 'but it's not plagiarism' crowd any favours for sure.
There's a significant difference between deliberately plagiarizing something and claiming that because you looked at something and remember what it looks like everything you do is plagiarism. Those people can continue to get fucked as far as I can see.
"Every word you used is in this book I have, that's publicly available and verifiable!"
-
@Arantor said in I, ChatGPT:
@DogsB the Twitter post I linked has some examples. In the interests of not making people go to Xitter I'll cross post.
There is an interesting argument being put forward that 'if I ask for a painting and I don't get something resembling the painting fairly closely, it's failed at giving me what I asked for' but there's no way to square that round peg without getting deep into the plagiarism stuff.
you don't charge an artist for having the Mona Lisa on their memory or using something they learned looking at it to paint an unrelated image
if you type "Mona Lisa" in the prompt it's your responsibility, the same way if you go to the original picture, that is publicly available, and make a similar one with your own hands
these anti-ai protests are dumb IMO
-
@sockpuppet7 the thing to note is that in previous versions of Midjourney, prompting Mona Lisa would get you something that was identifiably the Mona Lisa but was inevitably refracted through something else generative.
It was clearly 'based on' but held some distinct sense of identity, if that's a concept we can use.
This, however, fuels the fire of the plagiarism argument - because even an artist might be inspired by the Mona Lisa but they're not just going to go off and paint a good likeness from memory just because that was the prompt.
I do think there's an argument buried in there somewhere about making sure the art that's fed into these things is taken either from public domain or suitably licensed, because if not it does lead some interesting and dark places (like the knock-off Disney images that end up with the Disney logo in violation of Disney's trademark... real artists tend to avoid that particular pitfall)
-
@Arantor if your prompt is "Mickey mouse" and you use it, the lawyers at Disney will do the same than if you ask an artist to draw Mickey mouse and use it
even when I draw a barely recognizable Mickey mouse the result is the same
blocking the AI from training with public images of Mickey is akin of forbidding you from looking at it
-
@sockpuppet7 copyright laws do allow for things like parody and transformative works, not ones that are super close.
Artists can understand the difference and work with that as a guideline.
-
@Arantor said in I, ChatGPT:
@sockpuppet7 the thing to note is that in previous versions of Midjourney, prompting Mona Lisa would get you something that was identifiably the Mona Lisa but was inevitably refracted through something else generative.
It was clearly 'based on' but held some distinct sense of identity, if that's a concept we can use.
This, however, fuels the fire of the plagiarism argument - because even an artist might be inspired by the Mona Lisa but they're not just going to go off and paint a good likeness from memory just because that was the prompt.
Well...there's an argument that you can use it to produce copies of stuff. And you shouldn't do that and try to pass it off as yours or whatever.
I do think there's an argument buried in there somewhere about making sure the art that's fed into these things is taken either from public domain or suitably licensed, because if not it does lead some interesting and dark places (like the knock-off Disney images that end up with the Disney logo in violation of Disney's trademark... real artists tend to avoid that particular pitfall)
Well, there's nothing stopping you from taking the stylized artwork and then shopping the logos out. That shouldn't stop you from making Pixar-esque stuff just like nothing should stop you from drawing that stuff in the first place.
-
@Arantor said in I, ChatGPT:
@sockpuppet7 copyright laws do allow for things like parody and transformative works, not ones that are super close.
Artists can understand the difference and work with that as a guideline.
Right. And there's a person doing the prompting who's in the analogous position as that artist. Same same.
-
@boomzilla the difference is that the prompter might not think of dealing with such things, the artist likely will - and there's not really a good way to prompt to an AI to be sufficiently transformative to be original.
-
@Arantor said in I, ChatGPT:
@boomzilla the difference is that the prompter might not think of dealing with such things, the artist likely will - and there's not really a good way to prompt to an AI to be sufficiently transformative to be original.
Most prompters are just people fucking around. So who cares? If you're doing something commercial then you better figure shit out or you're going to find out. And that irregardless if it's AI art or anything else.
-
@boomzilla don’t discount the idiots however, there are a lot of them and the fumes are pretty high where they are.
-
@Arantor said in I, ChatGPT:
@boomzilla don’t discount the idiots however, there are a lot of them and the fumes are pretty high where they are.
I don't like the idea of slowing the pace of progress to cater for idiots
-
@sockpuppet7 said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@boomzilla don’t discount the idiots however, there are a lot of them and the fumes are pretty high where they are.
I don't like the idea of slowing the pace of progress to cater for idiots
This site's entire existence implies what happens if you don't.
-
@sockpuppet7 said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@boomzilla don’t discount the idiots however, there are a lot of them and the fumes are pretty high where they are.
I don't like the idea of slowing the pace of progress to cater for idiots
Just take the warning labels off. Let the idiots willingly do the rest.
-
@boomzilla said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@boomzilla the difference is that the prompter might not think of dealing with such things, the artist likely will - and there's not really a good way to prompt to an AI to be sufficiently transformative to be original.
Most prompters are just people fucking around. So who cares? If you're doing something commercial then you better figure shit out or you're going to find out. And that irregardless if it's AI art or anything else.
There’s a problem with putting the onus on the user.
Of course, the Mona Lisa is already in the public domain since forever, and everybody knows about the painting. So asking for pictures of her really is on the user’s side.
But what about all the other stuff it’s “learned”? If I ask an artist to make some painting for me and he plagiarizes something I’ve never seen or know anything about, it’s on the artist. I might have paid him, but he’s the one who did the plagiarism. If I ask the AI, which claims it doesn’t actually contain copyrighted works (which is basically disproven with this), to create some Mickey Mouse shit, I really should know better and the trouble is on me. But similar to the real artist above, if it plagiarizes something I don’t even know, it’s really the AI (actually, the people who trained it on stuff they don’t have the rights to) that’s to blame.Of course, none of this matters for most people just messing around with it.
-
@topspin said in I, ChatGPT:
@boomzilla said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@boomzilla the difference is that the prompter might not think of dealing with such things, the artist likely will - and there's not really a good way to prompt to an AI to be sufficiently transformative to be original.
Most prompters are just people fucking around. So who cares? If you're doing something commercial then you better figure shit out or you're going to find out. And that irregardless if it's AI art or anything else.
There’s a problem with putting the onus on the user.
Of course, the Mona Lisa is already in the public domain since forever, and everybody knows about the painting. So asking for pictures of her really is on the user’s side.
But what about all the other stuff it’s “learned”? If I ask an artist to make some painting for me and he plagiarizes something I’ve never seen or know anything about, it’s on the artist. I might have paid him, but he’s the one who did the plagiarism. If I ask the AI, which claims it doesn’t actually contain copyrighted works (which is basically disproven with this), to create some Mickey Mouse shit, I really should know better and the trouble is on me. But similar to the real artist above, if it plagiarizes something I don’t even know, it’s really the AI (actually, the people who trained it on stuff they don’t have the rights to) that’s to blame.Of course, none of this matters for most people just messing around with it.
Pretty sure he knows better than to explicitly ask for something specific. Because that's what we're talking about. The linked bruhaha was about people asking and getting the Mona Lisa.
Note that Disney isn't suing people for plagiarism (or anything else) generating Pixar-like movie posters. They just convinced MS to not let people use Bing to do it.
-
@topspin said in I, ChatGPT:
@boomzilla said in I, ChatGPT:
@Arantor said in I, ChatGPT:
@boomzilla the difference is that the prompter might not think of dealing with such things, the artist likely will - and there's not really a good way to prompt to an AI to be sufficiently transformative to be original.
Most prompters are just people fucking around. So who cares? If you're doing something commercial then you better figure shit out or you're going to find out. And that irregardless if it's AI art or anything else.
There’s a problem with putting the onus on the user.
Of course, the Mona Lisa is already in the public domain since forever, and everybody knows about the painting. So asking for pictures of her really is on the user’s side.
But what about all the other stuff it’s “learned”? If I ask an artist to make some painting for me and he plagiarizes something I’ve never seen or know anything about, it’s on the artist. I might have paid him, but he’s the one who did the plagiarism. If I ask the AI, which claims it doesn’t actually contain copyrighted works (which is basically disproven with this), to create some Mickey Mouse shit, I really should know better and the trouble is on me. But similar to the real artist above, if it plagiarizes something I don’t even know, it’s really the AI (actually, the people who trained it on stuff they don’t have the rights to) that’s to blame.Of course, none of this matters for most people just messing around with it.
the AI is only overfiting on images that have thousand of copies in publicly accessible pages, and it only makes copies for a very specific prompt naming it
there is no believable way ou can blame the tool if you create some infringing material on it
