Cipheron on 26/4/2023 at 01:18
I saw that "AI" generated pizza commercial and wasn't sure it's legit.
"Are you ready for best pizza of life?".
Straight up, that's NOT the type of thing ChatGPT would even output. It should be very rare to have the word sequence "for best".
The thing feels fake, because it includes grammatical errors to make it feel more "AI" but they're not even the type of errors that AI tend to make.
AI don't just randomly forget to have participles and pronouns. However that's common with some non-English speakers, whose native languages don't have those joining words.
So, a Russian speaker might conceivably say "Are you ready for best pizza of life?", but I can't see how the hell an AI would come up with that.
So it feels more like some human just removed words like "the" "of" "for" and "your" to make it feel more "robotic".
Tocky on 26/4/2023 at 20:16
An AI with access to worldwide language which it drew from perhaps? The images certainly align with how I've seen AI interpret them.
Cipheron on 27/4/2023 at 11:56
Well the guy says GPT4, so you'd assume that's pretty advanced.
Someone else actually called him out on the dialogue being too choppy to be GPT and the creator claims he deliberately asked GPT to write it poorly. Which is definitely possible.
demagogue on 27/4/2023 at 12:29
Yeah what's kind of funny about that is that, unlike a few years ago when those memes were going around (like "I trained an AI on 50 hours of Family Guy episodes and it wrote this scene") which pretty transparently were written by a person to sound like an AI, today ChatGPT is good enough that there's no reason to even bother with the subterfuge ... because you can just ask ChatGPT directly to sound like what people think cheap AI should sound like.
The catch I guess is that it's developing so quickly that a lot of people may still not even realize just how good it is, and they're still thinking about it like those old memes. So I could see it happening either way with this guy, either he actually did an oldschool style meme (by himself in its voice) without realizing ChatGPT is so good he didn't need to and he just came up with that story as a post-facto cover for his ignorance or whatever, or he knew how good it was and deliberately asked it to sound like the oldschool style meme. I actually don't care enough to get to the bottom of that mystery, only just enough to write this post. =V
Starker on 27/4/2023 at 13:27
My biggest worry about ChatGPT is that people are more and more starting to rely on it as a source of knowledge and problem-solving. Stack Overflow banned it several months ago already due to being overflooded with ChatGPT generated garbage answers and I've been seeing a similar reaction in language learning communities. It can generate useful and generally good enough explanations, but it's often subtly wrong in ways you can't really catch unless you already know the answer very well and the problems really start when you get to questions that aren't straightforward enough for them to be a good amount of text on it available on the internet or that haven't been prominently answered multiple times. This is an issue even very sophisticated algorithms like ChatGPT can't overcome -- it can only ever be as good as its source material, and usually it's less good.
WingedKagouti on 27/4/2023 at 15:32
Quote Posted by Starker
It can generate useful and generally good enough explanations, but it's often subtly wrong in ways you can't really catch unless you already know the answer very well
Systems like this can be used to give someone with at least moderate skill in an area ideas for alternative approaches, but low (and no) skill users will struggle to understand why the answers aren't useful on their own.
Starker on 27/4/2023 at 16:38
Sure, this is the best case scenario -- a user that already knows what they are doing and has plenty of skill to accomplish a task by themselves uses it to generate a boilerplate/mediocre solution that they can then independently verify is good/accurate enough. I'm not really worried about that especially since it has already been happening pre-ChatGPT, where people would use scripts/code written by other people and substitute their own variables or plagiarise an essay off the internet and change the wording slightly.
WingedKagouti on 27/4/2023 at 17:22
Quote Posted by Starker
Sure, this is the best case scenario -- a user that already knows what they are doing and has plenty of skill to accomplish a task by themselves uses it to generate a boilerplate/mediocre solution that they can then independently verify is good/accurate enough. I'm not really worried about that especially since it has already been happening pre-ChatGPT, where people would use scripts/code written by other people and substitute their own variables or plagiarise an essay off the internet and change the wording slightly.
The main issue with low/no skill people using solutions presented by ChatGPT (and similar) is trying to slot those solutions into something where they have potential to cause actual harm.
Moderately/highly skilled people using systems like this to improve their workflow is generally beneficial as long as they don't blindly trust what is offered.
Starker on 28/4/2023 at 04:32
There is also the problem that the sheer volume of algorithmically generated texts like these massively outweighs the number of people available to adequately assess these texts.
Also, from what I keep seeing, quite a few people seem to think that these texts are somehow created with some sort of an intelligence behind it (probably because it's being marketed as an AI), so they take them far more seriously and use them far more carelessly than they should.
I guess this must be what it was like when Eliza first appeared.
