I, ChatGPT
-
-
@Zerosquare said in I, ChatGPT:
@HardwareGeek said in I, ChatGPT:
I'm nominally a constant
#define HARDWAREGEEK 42READ ANOTHER BOOK!
-
-
@Zerosquare said in I, ChatGPT:
@HardwareGeek said in I, ChatGPT:
I'm nominally a constant
#define HARDWAREGEEK 42I blame autocarrot. Or AI.
Filed under:
-
Google scrambles to manually remove weird AI answers in search
/cdn.vox-cdn.com/uploads/chorus_asset/file/24440533/AI_Hands_A_Bernis_02.jpg)
Google’s AI Overview launch showcases that the race for AI domination is perilous.
Google Search’s “udm=14” trick lets you kill AI search for good
The power of URL parameters lets you unofficially turn off Google's AI Overview.
-
TIL, there is already a model that create images quickly on CPU only
GitHub - rupeshs/fastsdcpu: Fast stable diffusion on CPU
Fast stable diffusion on CPU. Contribute to rupeshs/fastsdcpu development by creating an account on GitHub.
(and a significantly smaller installation too)
-
-
-
@Zecc said in I, ChatGPT:
The trolling shitposts of yore comes back to haunt Google. It's beautiful!
-
@Carnage Well, at least it's not wrong this time.
-
@topspin said in I, ChatGPT:
@sockpuppet7 or you could just edit the HTML.
This. An LLM mispeling "homicidal" when it's spelled correctly in the prompt is way less likely than some
shitposter doing it.
-
@Arantor said in I, ChatGPT:
I wonder what they found.
I stumbled upon LLM Kryptonite and no one wants to fix it
Neural nets with flaws can be harmless … yet dangerous. So why are reports of problems being roundly ignored?
Customers buy software under the expectation that it will be maintained and supported. That's part of what they buy. The contract terms can vary, but at its essence, a customer purchases something under the expectation that it's going to work. And if it breaks – or doesn't work as promised – it will be fixed. If that doesn't happen, the customer has a solid grounds for a refund – possibly even restitution.
The "I haven't read license" thread is... wherever.
And yet another data point to my pet theory that the "as-is" license might be the greatest cancer of the software industry.
-
@remi as-is covers all of the open source licences that have any real meaning (because that’s the cornerstone of disclaiming liability through warranty)
You may believe that open source is also a cancer on the software industry but that is a different battle to fight.
I think there is a bigger problem than “as-is” though, which is those goddamn licenses. I believe quite firmly that if you buy a product and the product fails to work as advertised, or breaks through some defect that is not merely “normal wear and tear” you should be entitled to recompense or restitution, within reasonable grounds.
That’s not always possible, of course, but where it is, I believe we should normalise that. A piece of software doesn’t load a given (non-damaged) file properly? That’s a bug, especially if advertised as a feature that it does, and people should be able to get a patched version without paying for a new version.
Honestly this type of shit should have bankrupted Microsoft Office years ago with its failures to load even its own files correctly consistently, but we write it off as “shrug, computer says no” and then hand them money for the next iteration to fuck it up again.
Yes, the current world view of “getting what you paid for and having reasonable expectations of such” is a comedic matter, but it shouldn’t be, and it says a lot about us as a collective that doing the right thing is often both financially imprudent for companies and sometimes even the target of ridicule from peers. Why would you bother doing something well when you can half-arse it and make just as much money, plus selling the next version to the mugs again?
-
@Arantor said in I, ChatGPT:
I think there is a bigger problem than “as-is” though, which is those goddamn licenses. I believe quite firmly that if you buy a product and the product fails to work as advertised, or breaks through some defect that is not merely “normal wear and tear” you should be entitled to recompense or restitution, within reasonable grounds.
What you describe is what I call the "as-is" license, since in most cases the "as is" bit is part of the "warranty" section and immediately followed by "with no guarantee of performing any task, including the one you explicitly were told the software does perform" (that's not the legalese but it usually looks like it).
To wit, the GPL v3:
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM “AS IS” WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
There shouldn't be any need to shout but
to find a lowercase version to copy-paste.This is what I call "as is" and that I think is a very bad thing that should die (but never will).
I think we agree on that.
You may believe that open source is also a cancer on the software industry but that is a different battle to fight.
I don't, and it would be. But that's ir-
to my rant here.as-is covers all of the open source licences that have any real meaning (because that’s the cornerstone of disclaiming liability through warranty)
IANAL but I disagree. I think it would be possible to disclaim liability with some sort of "reasonable effort" formulation that would protect open source stuff (and other small actors).
Liability law for Real World objects (like, say, a household appliance blowing up) is, AFAIK, already full of this kind of "reasonable-ness" tests when things get to trials (was the use-case that caused the failure something that the manufacturer could have "reasonably" predicted, or did the manufacturer apply "reasonable" efforts to make their device behave in a correct way etc.), so I don't see why it couldn't be used in the software as well.
And yes this would make software very different, but that's the whole point of my rant.
-
@remi the problem is the notion of legal liability. Few open source projects that give any sort of shit will fail to deliver a best effort attempt to fix a problem. Indeed, many of them will go out of their way to try.
But they put that crap in their licences because businesses will use open source (because cheap) and want someone to sue when shit goes sideways because hitting the bottom line.
The bit you’re quoting of the GPLv3 is written by lawyers for lawyers. And explicitly is designed to shield hobbyists from being fucked over if something they made and gave away for free happens to fuck a business up (as the most egregious worst-case example).
For any such situation, open source will tend to get involved to try to fix it because it’s in their interests, business will tend to want not to as long as the cost of fixing fuckups remains below the cost of fixing the problem.
We agree that vendors across the board should do better. They should, they really should, but all the time it isn’t in their economic interests to do better, they won’t. As-is isn’t really the problem here.
-
Bestest title
-
@Arantor said in I, ChatGPT:
the problem is the notion of legal liability.
I don't really think, no. As I hinted in the previous post, the world of real physical objects hasn't got too many issues dealing with this. By which I mean it has tons of issues with it, but also tons of experience and legal practice and legal precedent that sets what's acceptable or not (roughly speaking, there always are and always will be edge cases).
You buy a fridge, it suddenly turns into an oven, company is liable. End of discussion. (they might refuse to pay and force you to be ruined in a trial or let the matter die and that's a shitty thing to do but that's not really the problem with liability)
Legal precedent (and probably law) has put a lot of safeguards around this, which I'm sure varies from place to place and lawyers so
but it's there.If you change the electronics of your fridge and it turns in an oven, tough luck. Or you sent the fridge down to the bottom of the ocean, fish it back and it turns into an oven, tough luck. Or it is 42 years old and an old rusted component turned it into an oven, tough luck. Etc.
If instead of buying a fridge from MegaCorp you buy a small gizmo from EtsyTinyBusiness, the same applies, with thresholds that are adapted through decades of experience.
Basically, it Works
. (more or less)If you're a business and you want a guarantee that EtsyTinyBusiness can't offer, either you accept that you will deal with it yourself, or you go find a larger business. Yes, it sort of sucks that the product of EtsyTinyBusiness can't be easily used by MegaCorp but that's also reflecting the actual reality of things where EtsyTinyBusiness simply can't really handle the needs of MegaCorp.
Now to software: if what I call the "as is" bit had never existed, then things would work like for Real World objects. And while things would be different (duh, again, that's my point), I don't see why it wouldn't work equally well (or bad...) than for anything else.
I am aware we don't live in a world where the "as is" bit never existed, which is (one of the reasons) why this is a rant. If a single company (or open source project) did the Right Thing (in my mind) and dropped it now, they would probably be fucked. So we're probably stuck with it forever.
But that still doesn't make this "as is" bit a good thing to me.
-
@remi ahhh I see your objection, and it’s not really the “as is” part. Because when you buy software now, you’re not getting it as-is, nor even as-was, you’re buying “whatever it becomes” which is very different and there aren’t legal protections against that shit yet, which there really should be.
I think the rise of SaaS is part of that problem where you’re not buying a thing and getting the thing for the life of the thing (where there are economic incentives to preserve the scope of features and minimise defects) but that it’s a constant stream of change.
You see this phenomenon elsewhere - gaming notoriously with its day one patch culture. But you also see the occasional success story - No Man’s Sky, arguably Fallout 76, arguably Cyberpunk 2077, these things changed drastically from what was sold day one, not just fixes of defects, and in virtually every respect for the better. But these are at best outliers.
I wish we could return to the era of software manufacturers operating much as they used to - they’d sell you Product X 1.0, and 1.0 through 1.0.x were fixes, and if you wanted the new features you could buy X 2.0 or you could… not… and just keep using 1.0.98 in perpetuity, such that developers tried where possible to fix bugs. (I know it was never quite that rose tinted. But it feels like it was better then, even without consumer protections to enforce it.)
-
@Arantor ... no, my objection hasn't got to do with the constant-update bullshit (though I also dislike that).
It really is the "as is" part, or rather the "without warranty of fitness for a particular purpose" (to reuse the GPL wording, some other licenses are even more straightforward in explicitly excluding the purpose you bought it for!) that I object to.
This, to me, is what enables software companies to put out shit, since they explicitly told you whatever you buy may not perform any function at all and it's not their fault. Whether they update their software every 5 min or every 5 years doesn't change that, if the software ends up formatting my disk, any software company can just shrug and walk away.
That is the thing that should never have existed (or been allowed to exist), but that we're sadly (IMO) stuck with now as it can't be undone.
-
@LaoC said in I, ChatGPT:
Bestest title
I keep reading "weird AI" as (lower-case-L, not upper-case-i). Definitely get Al Yankovic on that...
-
@remi I was going off the 'change the electronics of your fridge and it turns into an oven' part to infer the whole 'it's the constant change bullshit'.
Am I wrong to think software used to be less bad about this shit back in the day? It feels like it was less bad. But I also think software did less back in the day so it was clearer what you were getting and whether it would do or not do the thing you needed it to do.
-
@Arantor said in I, ChatGPT:
Am I wrong to think software used to be less bad about this shit back in the day?
Back then, post-release patching was expensive. That didn't mean there was no terrible and buggy software, but there was some incentive to do at least a minimum of QA before releasing. Also, that prevented "designers" from changing things and breaking features every fortnight.
-
@Zerosquare QA only matters if you care about quality though - there were shitty apps back in the day, just as there were unwinnable-due-to-bugs games back in the day.
It just feels radically worse now and about to accelerate.
-
@Arantor said in I, ChatGPT:
QA only matters if you care about quality though - there were shitty apps back in the day, just as there were unwinnable-due-to-bugs games back in the day.
Yeah, but releasing crap was a dangerous bet, because people who got a bad impression the first time were unlikely to be repeat customers for your company.
Nowadays, major video games publishers regularly ship hilariously broken stuff, and people just shrug.
-
There are jurisdictions in which
@remi said in I, ChatGPT:WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
doesn't satisfy
THE EXTENT PERMITTED BY APPLICABLE LAW.
When it's free software, you get what you paid for. But when someone is selling a program to do A, but the program you paid real money for does not in fact do A, there should be some liability for that, regardless of what their lawyers say.
-
@HardwareGeek alas, the legal wormlings have wormed their way out of that for the most part with those licences we were complaining about earlier.
-
@HardwareGeek said in I, ChatGPT:
When it's free software, you get what you paid for. But when someone is selling a program to do A, but the program you paid real money for does not in fact do A, there should be some liability for that, regardless of what their lawyers say.
I usually associate "as-is" licenses with free software. People are using that on paid software? That's hilarious.
-
@error said in I, ChatGPT:
I usually associate "as-is" licenses with free software. People are using that on paid software? That's hilarious.
Indeed, the "as-is" is "Look, I made this thing, it pretty much works for me, it might work for you, I don't know - no guarantees" and it's entirely up to you whether you use it or not. There's no contract being entered into.
But that changes once money starts changing hands.
-
@remi said in I, ChatGPT:
IANAL but I disagree. I think it would be possible to disclaim liability with some sort of "reasonable effort" formulation that would protect open source stuff (and other small actors).
For open-source stuff it's not a problem. When that breaks, you still get full refund (you paid nothing, you can have your nothing back anytime) and you get to keep all the pieces and can fix it either yourself or find someone who can if you consider it worth the effort.
It is the closed-source software where it is a problem. Because even when there is a support, you often find it's crap anyway.
-
@error said in I, ChatGPT:
@HardwareGeek said in I, ChatGPT:
When it's free software, you get what you paid for. But when someone is selling a program to do A, but the program you paid real money for does not in fact do A, there should be some liability for that, regardless of what their lawyers say.
I usually associate "as-is" licenses with free software. People are using that on paid software? That's hilarious.
Well, yes, most EULAs disclaim warranty.
-
@error said in I, ChatGPT:
I usually associate "as-is" licenses with free software. People are using that on paid software? That's hilarious.
To wit, I just picked the top result for "Microsoft Windows license text" and got this link, which looks like a reasonably recent Windows 10 license (and, for once, isn't longer than War and Peace!). Jump to section 9, in particular 9b.
"Microsoft [...] exclude all implied warranties and conditions, including those of merchantability, fitness for a particular purpose, and non-infringement."
I can't say I've actually read many licenses (that thread is... somewhere else) but I would go as far as saying that I don't think there are any software license, paid or not, that does not contain this kind of language.
@HardwareGeek said in I, ChatGPT:
There are jurisdictions in which ["as-is" isn't legal]
True, and for example the MS license above has a lot of wording to that effect. Which, btw, I find quite telling that a company feels the need to say "if the law says what we're doing is illegal, well, OK, sure, the law applies." But that's not specific to licenses, it's more the usual corporate greed and disrespect for rules.
To your point, it is possible that in the longer term laws are going to claw back some sanity from the insanity of "as is." Then again, since that's laws and politics, I redirect you to the
for the usual reaction to that, and how it works out in many cases. And it will probably take a long time until that cancer is properly eradicated through laws, if ever (probably never).@Bulb said in I, ChatGPT:
For open-source stuff it's not a problem. When that breaks, you still get full refund (you paid nothing, you can have your nothing back anytime) and you get to keep all the pieces and can fix it either yourself or find someone who can if you consider it worth the effort.
I disagree, because liability (in the real world of real objects) isn't limited to the value of what you bought but also the damages it caused. If your fridge turns in an oven and burns your house down, the fridge's manufacturer is liable for the price of your whole house, not just for the couple of hundreds bucks you forked for the fridge.
-
@Arantor said in I, ChatGPT:
Am I wrong to think software used to be less bad about this shit back in the day? It feels like it was less bad.
I think you're right but I'm not sure, there's a lot of bias in there.
For example, I remember the good games I used to play as a kid but I forgot the crappy ones. And there were some, I remember that once a friend gave us a large box of floppies (maybe 50 or so?), all games, and we only ever played a handful of those, the rest we deemed too crappy. I also was a kid so probably missed quite a few signs of crapiness even in games I did play. And if I liked a game I was likely to slog through the crapiness (and not necessarily remember it today, even if I noticed it back then) because I liked the rest.
@Zerosquare said in I, ChatGPT:
Yeah, but releasing crap was a dangerous bet, because people who got a bad impression the first time were unlikely to be repeat customers for your company.
True. In this big box of games there were at least a few that we started once, felt it was crappy in the first 2 min of play, and never ever touched it again. Granted, a lot of factors other than quality came into play (kids randomly discovering games...), but still.
Though OTOH, you paid for a lot of stuff where nowadays you can usually get a good taste of the software for free before paying anything. Demos did exist, yes, but that wasn't the rule. So once you had paid money to get the software, the developer (or rather, their accountant) didn't really care whether the software worked or not (within the limits that apply to any business...).
-
@remi sure there were always crappy games, that’s a given, but were many of them buggy? I find that an important distinction.
-
@Arantor I can't pinpoint an exact example from memory right now (*) but I'm pretty sure some games had "weird behaviour" that actually was a bug and that we just shrugged off. Some of them are still exploited today in e.g. speed runs. Not all bugs cause the game to become unplayable (or the software unusable, generally speaking). You learn to not do the thing that causes the crash, or that this feature doesn't work, and keep doing the rest.
Still, I fear there's too much bias in our memories, and to much that has changed in the software industry, to be able to do a proper comparison. I would tend to agree with you (that it was less buggy) but... I wouldn't swear to it.
(*) there's one that came back to mind while writing: in Worms, if you kept pressing simultaneously the F-keys for the minigun and the dynamite (?), and fired while doing so (you had to aim beforehand), it would fire a minigun burst where each shot was a dynamite. We tried all (?) other combination and only this one "worked." It clearly was a bug, but we lived with it and even enjoyed it, though that was considered "cheating" and the first one who did that turned a "serious" game into "let's blow the whole thing up and who cares who wins?"
-
@remi said in I, ChatGPT:
I think you're right but I'm not sure, there's a lot of bias in there.
The software was—by necessity of running on much less powerful computers—much simpler, and simpler software is easier to understand and less work to debug.
-
@Bulb yes, that most likely was a factor.
But conversely, there was much less use of libraries, frameworks and higher level languages. And, despite all the hate we love to pour on those, they also work in avoiding some classes of bugs (they create new ones, but remove some others). Stuff like buffer overflows was likely hugely more prevalent when everyone had to reinvent the wheel and code very close to the metal.
I guess overall, so much has changed in software since then (and btw, when exactly was "then?") that we're simply not comparing apples to apples (inb4: joke).
It's like trying to decide if a Ford T was more or less reliable than a modern car. There's so much different, both in what the car does, how it operates, and what we expect from it that the comparison makes little sense.
-
One thing that is improving but is still hard to make something like a full book on AI is the limit of the size it can write while being coherent. If I put too much context, it "forgets" a lot of stuff and makes a mess. Then a common strategy is to divide it into chapters. I ask it for the outline for a book, then, now write me chapter 1
It starts well, but then ends the story on the same request. GPT-4o did improve the length it can make in one go, I can convince it to output around 2000 words, but has the same problem.
If I take it take something like this: this is the outline, write from the part where Frodo took the ring, and write until Gandalf fights the Balrog, it would probably start well, suddenly the hobbits solves several problems and battles and the ring is melt at Mordor
The super-hyper large context some recent models claim doesn't work on this use case.
I guess it's probably possible to automate it making the LLM itself query the last paragraphs and check for these slips and etc
But who wants to write a lot of code attempting to do it well just to be surpassed for some next version of these LLM?
-
@DogsB That thread is over
-
Status: I saw that Google wanted to translate someone's comment on YouTube...
Seems like that's not English enough. Let's see what it thinks...
-
@Tsaukpaetra said in I, ChatGPT:
I saw that Google wanted to translate someone's comment on YouTube...
Yeah, I've been seeing that a lot. Google has started sticking that that on any comment that contains laughter or anything else that isn't an actual word.
-
-
@sockpuppet7 said in I, ChatGPT:
One thing that is improving but is still hard to make something like a full book on AI is the limit of the size it can write while being coherent. If I put too much context, it "forgets" a lot of stuff and makes a mess. Then a common strategy is to divide it into chapters. I ask it for the outline for a book, then, now write me chapter 1
It starts well, but then ends the story on the same request. GPT-4o did improve the length it can make in one go, I can convince it to output around 2000 words, but has the same problem.
If I take it take something like this: this is the outline, write from the part where Frodo took the ring, and write until Gandalf fights the Balrog, it would probably start well, suddenly the hobbits solves several problems and battles and the ring is melt at Mordor
The super-hyper large context some recent models claim doesn't work on this use case.
I guess it's probably possible to automate it making the LLM itself query the last paragraphs and check for these slips and etc
But who wants to write a lot of code attempting to do it well just to be surpassed for some next version of these LLM?
Something, something, initial conditions...
-
https://openai.com/index/openai-board-forms-safety-and-security-committee/
Today, the OpenAI Board formed a Safety and Security Committee led by directors Bret Taylor (Chair), Adam D’Angelo, Nicole Seligman, and Sam Altman (CEO)
They will investigate themselves and will find nothing wrong.
At the conclusion of the 90 days, the Safety and Security Committee will share their recommendations with the full Board
Oh, sorry. They will have some more meetings about things they already want to do, to find that they agree that the things they want to do should be done.
-
@Applied-Mediocrity of which the circles will recommend legislation that lets them carry on but prevents future players joining the market.
-
@Arantor Not that legislation would stop anyone blowing the big bubble, but yes.
-
@Applied-Mediocrity said in I, ChatGPT:
https://openai.com/index/openai-board-forms-safety-and-security-committee/
Today, the OpenAI Board formed a Safety and Security Committee led by directors Bret Taylor (Chair), Adam D’Angelo, Nicole Seligman, and Sam Altman (CEO)
At the conclusion of the 90 days, the Safety and Security Committee will share their recommendations with the full Board
And their conclusion is:
-
Week 3 of Salesforce training. We're talking about chatbots. It's being aggressively pitched.
I just heard GenAI referred to as "true" AI, which to me further muddies the concept of AI vs AGI.
-
@error said in I, ChatGPT:
I just heard GenAI referred to as "true" AI, which to me further muddies the concept of AI vs AGI.
It must be The No True
ScotsmanAI Fallacy in action...
-
@error said in I, ChatGPT:
Week 3 of Salesforce training. We're talking about chatbots. It's being aggressively pitched.
Of course it is, it's a paid feature.
-
Someone at work just posted to the technical chat that there's a script that some people might not be aware of (I wasn't) to simplify creating new
$languagefiles. Rather than starting with an empty text file, the script fills in boilerplate like the company copyright and confidentiality text. You can specify what kind of file you are creating (module, interface, test, etc.), and it will add the appropriate boilerplate for that.Someone else asked (humorously — I think) if the script author could add AI to the script and have it just write the entire file from the spec.
