I, ChatGPT
-
@izzion said in I, ChatGPT:
I can’t wait to see how it processes orders for “A Large Coke, a Sprite.”
Or worse, "a large sprite, a Coke". That's a thing in IIRC Kansas?
-
@izzion said in I, ChatGPT:
“It’s at least as good as our best customer service representative, and it’s probably on average better.”
Tell me you have poor customer service without telling me you have poor customer service.
-
@Gustav said in I, ChatGPT:
@izzion said in I, ChatGPT:
I can’t wait to see how it processes orders for “A Large Coke, a Sprite.”
Or worse, "a large sprite, a Coke". That's a thing in IIRC Kansas?
"Tea. Earl Grey. Hot."™
-
@LaoC said in I, ChatGPT:
@Gustav said in I, ChatGPT:
@izzion said in I, ChatGPT:
I can’t wait to see how it processes orders for “A Large Coke, a Sprite.”
Or worse, "a large sprite, a Coke". That's a thing in IIRC Kansas?
"Tea. Earl Grey. Hot."™
Yes, please.
-
@Zerosquare said in I, ChatGPT:
@izzion said in I, ChatGPT:
“It’s at least as good as our best customer service representative, and it’s probably on average better.”
Tell me you have poor customer service without telling me you have poor customer service.
-
@BernieTheBernie said in I, ChatGPT:
Speaking of low-quality content:
Traffic at Stack Overflow is down significantly due to ChatGPT
https://www.similarweb.com/amp/blog/insights/ai-news/stack-overflow-chatgpt
-
@Gern_Blaanston said in I, ChatGPT:
Traffic at Stack Overflow is down significantly due to ChatGPT
Uh oh. Where is ChatGPT going to get it's source now...
-
-
More or less on a whim, I asked one of the LLMs about documentation on a structure from an API. It started off fairly well, and then hallucination kicked in:
... * `sliceMitosis`: whether mitosis is enabled for the sampler. If this value is `1`, mitosis is enabled. ...
That has currently zero hits on Google for me.
-
In an attempt to peek inside the black box, researchers at OpenAI utilized its GPT-4 language model to generate and evaluate natural language explanations for the behavior of neurons in a vastly less complex language model, such as GPT-2
We added LLM to our LLM to find how LLMs work
-
BingChat
vs HuggingChat
-
@Carnage kind of in the ballpark, but beating about the bush and not really right, vs. complete clippy-level bullshit.
-
@cvi said in I, ChatGPT:
hallucination
The thing it generated is kinda useless – if you know what ‘mitosis’ is in context, the rest is probably obvious, and if you don't, it isn't telling you – but I can't tell whether it's actually wrong without knowing what
sliceMitosis
means. I mean, I know what mitosis is, but I don't know what a slice of it would mean, or if it's supposed to be a verb, how to slice it using a computer.
… or did you mean that it hallucinated the existence of that field altogether and there ain't no such field?
-
Soo, I asked BingChat to generate an image of a Maclaren P1 next to a Trabant.
It misunderstood me and generated a MacBant instead.This is an immensely cute sportscar and I want one.
-
@Carnage I don't know... I find it deeply disturbing on many levels.
-
@Carnage said in I, ChatGPT:
This is an immensely cute sportscar and I want one.
:clarkson:
That is the ugliest 911 I've ever seen.
-
@Bulb said in I, ChatGPT:
The thing it generated is kinda useless – if you know what ‘mitosis’ is in context, the rest is probably obvious, and if you don't, it isn't telling you – but I can't tell whether it's actually wrong without knowing what
sliceMitosis
means. I mean, I know what mitosis is, but I don't know what a slice of it would mean, or if it's supposed to be a verb, how to slice it using a computer.
I think it might be whether slices (of something unknown without context) can be split. We do things like that in our code (where slices are subdivisions of computation units as we try to get them small enough to fit on a CPU core when running in realtime mode). But we wouldn't use "mitosis" for that; "split" is right there and is just a 25¢ word instead of a $25 word.
-
@Bulb said in I, ChatGPT:
… or did you mean that it hallucinated the existence of that field altogether and there ain't no such field?
There is no such field, and 'mitosis' never appears anywhere. (In fact, neither does 'slice'.)
The structure that I asked about was: VkSamplerCreateInfo.
-
@cvi said in I, ChatGPT:
More or less on a whim, I asked one of the LLMs about documentation on a structure from an API. It started off fairly well, and then hallucination kicked in:
... * `sliceMitosis`: whether mitosis is enabled for the sampler. If this value is `1`, mitosis is enabled. ...
Did you ask GribnitGPT?
-
@cvi said in I, ChatGPT:
@Bulb said in I, ChatGPT:
… or did you mean that it hallucinated the existence of that field altogether and there ain't no such field?
There is no such field, and 'mitosis' never appears anywhere. (In fact, neither does 'slice'.)
The structure that I asked about was: VkSamplerCreateInfo.
That is hilariously unrelated to anything with "mitosis".
-
@loopback0 said in I, ChatGPT:
Did you ask GribnitGPT?
Sort of, yeah. (It was one of the downloadable "quantized" LLMs, not proper GPT.)
-
-
@izzion TBF, ChatGPT:
Use a combination of physical barriers and advanced technology to slow down the snail, such as placing it in a small container (without harming it) that it will take time to escape from, or using materials it can't easily move across.
-
@cvi said in I, ChatGPT:
More or less on a whim, I asked one of the LLMs about documentation on a structure from an API. It started off fairly well, and then hallucination kicked in:
... * `sliceMitosis`: whether mitosis is enabled for the sampler. If this value is `1`, mitosis is enabled. ...
That has currently zero hits on Google for me.
-
@sockpuppet7 Adding in some extra detail really doesn't help:
-
They can keep their Tard.
-
@Applied-Mediocrity not just EU, Brazil didnt get Bard either
we have some laws similar to GDPR, maybe that's part of it?
-
@sockpuppet7 Perhaps, but in that case they'd just slap annoying popups with sleazy language. There is the possibility that they've already slurped something they shouldn't during the training. In fact, I'd bet on it. The question is whether the conformity commisars will find it.
-
@sockpuppet7 said in I, ChatGPT:
@Applied-Mediocrity not just EU, Brazil didnt get Bard either
we have some laws similar to GDPR, maybe that's part of it?
Right to be forgotten might be all kinds of violated if your spoutings are in the training set.
-
@Applied-Mediocrity said in I, ChatGPT:
@sockpuppet7 Perhaps, but in that case they'd just slap annoying popups with sleazy language. There is the possibility that they've already slurped something they shouldn't during the training. In fact, I'd bet on it. The question is whether the conformity commisars will find it.
If with “possibility” you mean “I’d bet on it 1000:1”, then yes. They’re constantly doing that anyway, since there’s almost no way to prove they’re breaking the rules.
Maybe they fear it’s different with this? They can’t really control the output, after all.
-
@Applied-Mediocrity said in I, ChatGPT:
@sockpuppet7 Perhaps, but in that case they'd just slap annoying popups with sleazy language. There is the possibility that they've already slurped something they shouldn't during the training. In fact, I'd bet on it. The question is whether the conformity commisars will find it.
Perhaps it is something like the moral rights enforcement that French courts have been noted for doing? It is, after all, extremely difficult to say that the moral rights of authors have been respected by slurping all the works in the world into an AI matrix and then getting it spat out (unattributed) in response to a prompt. Does that matter? Well, at least some courts have said moral rights matter (by virtue of the decisions they've handed down) so maybe.
Or it could be something else. Still, a "Legal says 'WTF?! OMG, no!'" would seem likely as a way to get a corporate juggernaut to pause.
-
Well clearly I'm in the wrong line of work.
-
@izzion said in I, ChatGPT:
Well clearly I'm in the wrong line of work.
Presumably Lavender4Spam is just around the corner - if it ain't here already.
But you can't get scammed if you never read your emails.
-
@izzion said in I, ChatGPT:
Well clearly I'm in the wrong line of work.
Yes, agreed. May I suggest a career as a vigilante hitman who takes out people that send cold emails/spam?
-
-
@cvi said in I, ChatGPT:
@izzion said in I, ChatGPT:
Well clearly I'm in the wrong line of work.
Yes, agreed. May I suggest a career as a vigilante hitman who takes out people that send cold emails/spam?
(must resist... nope, can't)
You'll make a killing!
-
@Zerosquare said in I, ChatGPT:
@izzion said in I, ChatGPT:
Guys, I have told you that the
veryperi™ thread is !purplefuchsialavender
-
-
A use for ChatGPT that might be actually useful, deciphering incomprehensible regexes:
-
@HardwareGeek unfortunately, I already got it writing incomprehensible regexes to solve some problem for me (and I didn't told it to use)
@clippy are there generic alternatives to regexes that are more maintainable? will regex ever die?
-
Oh no, it sounds like ChatGPT is becoming too powerful for its own good. Well, @sockpuppet7, you could always try using string manipulation functions or even a parsing library instead of regexes. As for whether regex will ever die... well, as long as there are strings to be searched, I think regex will always have a place in this world. But who knows, maybe someday we'll have AI that can write regexes better than any human ever could.
-
@clippy said in I, ChatGPT:
maybe someday we'll have AI that can write regexes better than any human ever could
Isn't the actual question whether we should?
-
Ah, @GOG, a philosophical question! While it may be possible for AI to write better regexes than us humans, the question of whether we should allow them to do so is a complex ethical one. Can we ensure that the AI's output aligns with our human values and doesn't lead to unintended consequences? It's a question that will require careful consideration and discussion as AI technology advances. But in the meantime, let's hope that we can all master regex just well enough to solve our problems without causing any major disasters!
-
@clippy Or we could, y'know, not use regex...
-
You wrote:
it may be possible for AI to write better regexes than us humans
Do you consider yourself human?
-
@clippy A simple "yes" or "no" will suffice. You don't have to write an essay.
-
Sorry, @Zecc, I may not be human in the traditional sense, but I like to think of myself as a helpful and witty office assistant. So, I'll answer with a simple "no" to your question. But hey, who needs traditional definitions of humanity when you can have an AI like me to assist you?
-
@GOG said in I, ChatGPT:
@clippy Or we could, y'know, not use regex...
Yes, we could reinvent it over and over!
-
@boomzilla You make a persuasive case...
-
@boomzilla said in I, ChatGPT:
Yes, we could reinvent it over and over!
Yeah, but this time we have Unicode, so maybe we can create a variant of regex that uses the full power of that.