I, ChatGPT
-
On X, University of Washington computer science professor (and frequent OpenAI critic) Pedro Domingos wrote,
"Ilya Sutskever's new company is guaranteed to succeed, because superintelligence that is never achieved is guaranteed to be safe."
-
-
-
@topspin And memfrob is similarly useless since it xor's with 42. ASCII stays ASCII, the only letter which becomes nonprintable is
U
(sent to DEL), and though the whole 32-63 codepoint range goes to control characters specifically space goes to newline (whichstrings
does print).
-
-
-
-
-
they're aiming high now, we'll either have a laugh or a very interesting AI, or both
-
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
-
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
Don’t bad mouth all those PhDs in underwater basket weaving
-
@sockpuppet7 said in I, ChatGPT:
this is cool
I wonder if it will consistently camel-toe the little girls that don't have very obvious covering, like in the given examples....
Not that I'm complaining, just that you don't usually see that in underdeveloped people, you know?
-
@izzion said in I, ChatGPT:
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
Don’t bad mouth all those PhDs in underwater basket weaving
They can do whatever they want with their mouths while weaving underwater.
-
@sockpuppet7 said in I, ChatGPT:
they're aiming high now, we'll either have a laugh or a very interesting AI, or both
PhD: Piled Higher and Deeper.
If they are that sure of themselves, they should say that they'll have it authoring a thesis (without unacknowledged human assistance) and passing a public viva in a set number of years. Say, 3 or 4? If it manages that, it will be able to genuinely claim a PhD. People have the same sort of rules.
But if they haven't got the system doing reasoning and retaining a long context, it surely won't pass. Vivas can be quite antagonistic at times.
-
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
In case of Gender Studies, Sociology, etc. nobody will find out.
-
@dkf Eh, if it aims for a PhD in AI stuff, it might pull it off. Enough seem to go through their PhDs with regurgitating a few existing networks in different combinations. Sounds AI:able.
-
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
Don't knock it. It's writing most of the papers at the moment.
Oddly enough, I thought this would be a good use of LLMs. Feed it all the current literature and see what odd connections it makes.
-
@DogsB said in I, ChatGPT:
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
Don't knock it. It's writing most of the papers at the moment.
Oddly enough, I thought this would be a good use of LLMs. Feed it all the current literature and see what odd connections it makes.
If I could focus on the fun parts of science and let large language models handle the bureaucracy, formatting, and other tedious tasks, I would have studied longer
-
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
There's an education thread that way
-
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
-
@BernieTheBernie said in I, ChatGPT:
@topspin said in I, ChatGPT:
@sockpuppet7 yeah, PhD in what?
Let’s just wait for the science it publishes.
In case of Gender Studies, Sociology, etc. nobody will find out.
@error_bot xkcd impostor
-
-
@Tsaukpaetra said in I, ChatGPT:
@sockpuppet7 said in I, ChatGPT:
this is cool
I wonder if it will consistently camel-toe the little girls that don't have very obvious covering, like in the given examples....
Not that I'm complaining, just that you don't usually see that in underdeveloped people, you know?
the model in the setup is one of those fine tuned sd biased into sexy girls. with the test I did so far I got better results with sd base model
there are fine tunes that aren't like that, but these are underused, and the popular ones are merges of everything under the sun
-
-
There's indeed been a lot of delving lately.
-
-
-
@DogsB said in I, ChatGPT:
-
Have you tried asking ChatGPT about bypassing the paywall?
-
@error It's the telefrag. It has a well known workaround for the paywall—switch to the Firefox reading mode as soon as the icon comes up in the url-bar. It might also be possible with noscript.
-
llama.ttf is a font file which is also a large language model and an inference engine for that model.
Of course.
Reminds me of @blakeyrat's favorite book, A Fire Upon the Deep, where they talk about stuff like "sentient network protocols." It all sounded appropriately weird (but plausible) in the book. I guess it's still weird. Just more realistically plausible that some madman would make something like that.
-
@boomzilla That's really!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
-
-
I just tested, and there was a large difference on the experience of asking gpt-4o to write typescript and javascript. Strongly typed languages are the way to go if you want to gpt it.
I got this written by gpt-4o, it's a simple thingy: https://github.com/machado2/sdcompare
It wasn't a single gpt question, it took me a few hours. It was probably slower than writing it by hand if I was in the mood, but it was less boring, and the way
-
@sockpuppet7 said in I, ChatGPT:
Strongly typed languages are the way to go
if you want to gpt itFTFY
-
@topspin said in I, ChatGPT:
@sockpuppet7 said in I, ChatGPT:
Strongly typed languages are the way to go
if you want to gpt itFTFY
I agree, but I can never make my mind on programming the stack to use. I wrote the wibble on typescript with next.js, then I rewrote the backend on python, then I rewrote everything into rust with no javascript, then I started, but never finished, to move it to all python.
While I'm wasting time with this, @Arantor stayed firm on PHP, and while we're laughing at PHP he got GPT-famous, 2 wifes, and probably a billion USD yatch
-
@sockpuppet7 said in I, ChatGPT:
rewrote everything into rust
-
@Applied-Mediocrity said in I, ChatGPT:
@sockpuppet7 said in I, ChatGPT:
rewrote everything into rust
I keep feeling that with enough practice I would be as productive with rust than with other high level languages, but I'm not yet there
I felt it's more work when I need to change things too, that's a big problem
still better than c++ if you need the low levelness IMO
-
Yeah, this shit has been around for a while.
-
@sockpuppet7 said in I, ChatGPT:
I keep feeling that with enough practice I would be as productive with rust than with other high level languages, but I'm not yet there
I don't actually think so. There still is more things to mind in Rust than in Python or TypeScript. All the references, Arcs, Cows, Pins and the like are going to always get in your way and slow you down. Language that just has a
str
will always be simpler to write than a language that has an&str
,String
andCow<str>
.That said, the features currently unique to Rust make managing complexity simpler, so the complexity may pay off for large or very complicated projects.
I felt it's more work when I need to change things too, that's a big problem
This again very much depends on the complexity. When you understand the code well, the additional compiler checks are not doing much help for you, but you still have to tweak things all over the place to make them pass.
But when you have code that's been worked on for a couple of years by many people and you have little idea what large parts of it do, those checks will allow you to not treat it as a shitted stick, because you will be confident that if the borrow checker isn't complaining, the thing you are changing really isn't used anywhere else.
still better than c++ if you need the low levelness IMO
Definitely. I would however also note that C++ written 30 years ago and C++ written 10 years ago using modern techniques are very different in their maintainability.
-
@Bulb said in I, ChatGPT:
Definitely. I would however also note that C++ written 30 years ago and C++ written 10 years ago using modern techniques are very different in their maintainability.
I have coworkers that code C++ like it's 1990. At least with something that didn't exist at the 90s has that advantage
-
@Bulb said in I, ChatGPT:
But when you have code that's been worked on for a couple of years by many people and you have little idea what large parts of it do, those checks will allow you to not treat it as a shitted stick, because you will be confident that if the borrow checker isn't complaining, the thing you are changing really isn't used anywhere else.
I'd be afraid a few years of mediocre developers working around the borrow checker yields worse (even though safer) code than if the project had been in C++.
-
@PleegWat said in I, ChatGPT:
@Bulb said in I, ChatGPT:
But when you have code that's been worked on for a couple of years by many people and you have little idea what large parts of it do, those checks will allow you to not treat it as a shitted stick, because you will be confident that if the borrow checker isn't complaining, the thing you are changing really isn't used anywhere else.
I'd be afraid a few years of mediocre developers working around the borrow checker yields worse (even though safer) code than if the project had been in C++.
Everything will become an Arc and leak memory on cycles
-
Huh...I've never seen that...
Notably, the label is only visible on the mobile apps and not on the web.
-
@boomzilla this is a stupid idea anyway, what if a photographer does a small tweak with AI? how is that different from photoshop? what about photoshop context-aware thingies?
-
Test your recognition skillz:
https://www.pcmag.com/articles/how-to-detect-ai-created-images
-
-
-
From the authors of Crypto collapse? Get in loser, we’re pivoting to AI and The LLM is for spam comes Pivot to AI: the site
-
@Applied-Mediocrity said in I, ChatGPT:
Test your recognition skillz:
https://www.pcmag.com/articles/how-to-detect-ai-created-images
I got 10 of 12 correct.
The two I got wrong were supposed to be photographs taken on the street. And there were some more supposed photographs that I wasn't sure about.
The art ones were relatively easy, because the AI images were either too crisp and clean, or didn't match the style of the supposed artist. In the first one I was quite sure the simple one does fit Picasso's late style and since “both” wasn't an option, the detailed one had to be the generated one.
Also one question (the cow) became trivial when I noticed they left the Craion watermark in it.