ChatGPT

[Ari]

I find it interesting that it went for the pure color for white, gray, blue, red and one of the two yellows but green is at half value, which kinda shows how we’re extremely sensitive to green. Full red is red, full blue is blue but full green is neon/lime.

I do have to wonder if asking it in hex codes without any detailed language is skewing the color palette towards good old fashion web safe/standard colors. Also I wonder how the image generation networks would handle being asked to generate color swatches and then chatgpt told to pull hex data from those swatches.

[erimir]

I've actually been working on a project in this space for the past few months... Although I suppose anything interesting I'd have to say based on my experience would be probably something I'm not supposed to say.

[JoeP]

Originally Posted by erimir

Although I suppose anything interesting I'd have to say based on my experience would be probably something I'm not supposed to say.

It's all right, no one links here

[specious_reasons]

Originally Posted by JoeP

It's all right, no one links here

You can post classified documents to impress us, we don't mind.

[Kamilah Hauptmann]

Originally Posted by erimir

I've actually been working on a project in this space for the past few months... Although I suppose anything interesting I'd have to say based on my experience would be probably something I'm not supposed to say.

Ah, a War Thunder player

[Kamilah Hauptmann]

Saturday Morning Breakfast Cereal - Muffins

[Pete Ross Junior]

Tried to get into ChatGPT and it asked me to prove I'm human.

Too good for their own kind?

[-FX-]

Did you pass the test or not?

[Pete Ross Junior]

I got a human I know to help me out.

[-FX-]

Prove it

[LarsMac]

Never asked me to prove anything. It knows me, I reckon.

[-FX-]

It does bring up a real problem. Chatbots talking to each other. Considering the speed at which they respond, it could end up using all bandwidth worldwide in a matter of seconds.

And filling up all the servers.

[erimir]

I can share that I'm not working on Elon Musk's ChatGPT un-woke competitor

[-FX-]

An AI (actually just a large language model), with no censorship or bias, would scare everybody. Except the extreme radicals that want everything out there, and let the chips land where they will.

Free speech loving anarchists would embrace an actual AI. The rest of us would kill it with fire.

[LarsMac]

Originally Posted by -FX-

It does bring up a real problem. Chatbots talking to each other. Considering the speed at which they respond, it could end up using all bandwidth worldwide in a matter of seconds.

And filling up all the servers.

Eh, they'd run out of shit to talk about pretty quick.

[Ari]

Originally Posted by -FX-

Free speech loving anarchists would embrace an actual AI.

That’s what I’m worried about, that people who love to label themselves “Free speech loving anarchists” or “Free speech absolutist” will see the Mirror that is ChatGPT saying things they agree with and declare it an unbiased look at reality.

I think a lot of humans are about to fail the mirror test and presume the thing looking back at them in the mirror is another creature and not just themselves reflected back in a way they’ve rarely seen before.

[LarsMac]

I've actually never met a "Free Speech loving Anarchist."

[fragment]

Originally Posted by -FX-

An AI (actually just a large language model), with no censorship or bias

If you trained it on text without any kind of selection criteria you'll just get something that tries to sell boner-enhancing Ponzi schemes.

[Pete Ross Junior]

Originally Posted by -FX-

... The rest of us would kill it with fire.

I'm afraid I can't let you do that/

[-FX-]

Originally Posted by LarsMac

Eh, they'd run out of shit to talk about pretty quick.

Based on how the social media sites work, they would not only never run out of shit to talk about, they would also makes shit up.

[Crumb]

You are a silly person.

[Ensign Steve]

Once again, Chat GPT comes through for me in my job.

I asked it to write a JSON parser in modern C++ which maintains the object map keys in insertion order (jsoncpp does not do this, and there's a patch for it that does, but it sucks). The robot produced a bunch of placeholder code with the necessary function signatures, but with TODOs for all the implementations (lol just like a real engineer!). That was actually pretty helpful of it to at least generate a good starting place, but that's not even the best part.

ChatGPT said (correctly albeit condescendingly):

Please keep in mind that implementing a JSON parser from scratch is a complex task. Relying on an existing library, such as nlohmann::json, is advised when possible, since it provides a tested, efficient, and standards-compliant JSON parsing and serialization.

nlohmann::json you say? Yes! It's fucking perfect.

That's twice now it has answered me with poor/incomplete code while at the same time making me aware of an existing library that does exactly what I need.

[Kevlar]

So, I do these presentations at my company for all the developers and my next one was going to be ChatGPT. When my bosses found out they shit their pants and came at me like "you can't talk about that until we have a security policy!". I responded that we should blacklist it at work but people will still use it at home. They shit all over that idea too. They are really scared of it. I wish they'd replace management with ChatGPT. At least it would be 80% competent.

[Ensign Steve]

How can we craft a security policy without talking about it?

[Pyrrho]

Where I work, people are competing for prizes for ways to use it.