Joshua Ellis

Donald Trump Announces He Is The Recipient Of A "Major Award"

Fifth iteration of Samurai Capybara is going way more Kid Robot style. Easier to articulate and print and design accessories for.

The first iteration looked cool but just printed like absolute shit - his skirt thing was physics modeled and very clever but vanished when turned into printable geometry. Belt too. You have go big when designing this stuff, I think.

His swords didn't print properly at all - too thin and the handles separated for some reason. But hey, a week ago he didn't exist even as an idea and now he's a real thing in the world. My first attempt at designing a 3D printed character instead of functional parts.

My Samurai Capybara has gone from the virtual world to the real one! He's not perfect and I think I'm gonna do one more iteration that's more stylized and "Kid Robot vinyl art toy" style. But I'm really proud of creating this from nothing - no AI, just Blender modeling and sculpting.

Why a samurai capybara? Why not? 😂

A 3D rendered image of a samurai capybara

I've decided to design and print my own art toys for fun, and this is my first real attempt in Blender. I am stupid proud of this for a first effort. I've gotta fix some of his geometry but he'll be awesome to print and paint. Next step: joint articulation!

@ghostpanther.bsky.social hey, man, did you email me personally today? If not, someone is running a scam pretending to be your company to get people to pay for a workshop thing. Figured you'd want to know.

Got my Halloween costume sorted out

yeah

I don't ask ChatGPT to solve my life problems for me, I ask it to write code. It's a good and reasonable use case for it, it's not stealing anyone's soul. I asked people to help me and they didn't. It's not a big ask.

Probably should, if it's a UK number.

Yes! (Sorry, I don't check in here much rn.) I use Ubuntu 24.05 on a convertible Dell laptop/tablet and Raspberry Pi OS on a couple of different Pis, all with customized interfaces because I'm a big nerd. My daily driver is a MacBook Pro, though.

Remember, I literally uploaded a zip of my entire API routes folder like two hours before I started catching on to all this. I then went to the local cafe to get coffee and read for an hour and a half, came back, and it had no memory of this.

That's bizarre, right? It's not just me, right?

And just in case you think my expectations are unreasonable: when you do the paid subscription you get access to "Projects" where you can "add files" and such. It even says on my Kowloon Project in the ChatGPT app that I have "1 file attached".

Then look at my last interaction with the AI itself.

So if anyone can explain to me why I shouldn't be very angry about this waste of time and money or why I shouldn't cancel my subscription right now and delete the app and just resign myself to spending months more than I'd thought on my project, please, I am all fucking ears.

As it says: this is a deliberate choice on the part of its owners, as is the choice to not make their customers aware of this beyond the "hahaha, this thing makes shit up, don't trust it" disclaimer.

But it's not making up bullshit, it's getting its hippocampus wiped every few minutes.

Again: I don't anthropomorphize AI and I don't use it to be my pal or my therapist. I have one thing I mostly use it for that is supposed to be the one thing it's better at than anything else, except bullshitting. And it turns out it can't even do that.

But it's not a technical limitation, is it?

I don't think it was unreasonable of me to assume that if I'm paying for a tool and it allows me to upload files to a "project" folder, that it would not forget that information as soon as I went to take an extra long piss or get coffee.

Computers usually, and let me be clear, DON'T FORGET SHIT.

I am trying to imagine what it would be like to collaborate with an experienced dev with Guy Pearce/Memento style short term amnesia, and... yep, it would be like this. And I'm paying for this. (Though not for much longer.)

And it has no self-awareness so it can't even *tell* me, which is like weaponized pathos. It just happily agrees with anything I say even though it literally has no memory of what the topic of conversation is.

This is not very good when you're using it as a development partner/tool.

If I close a chat window, or go away from the chat for an unknown amount of time, or if OpenAI runs a load balancing switch, or for no apparent reason at all, it instantly forgets *everything I've told it* except for basic preferences. It forgets what I wrote, what it wrote, the code I gave it....

But this week it started acting exactly like people I've known with dementia -- forgetting stuff we'd literally talked about a half hour earlier, forgetting its own code, imagining we'd used libraries I'd never even heard of, contradicting itself. I couldn't figure it out.

Then I asked outright.

I do not think ChatGPT is sentient and I understand it's a Markov chain on bathtub crank, but I thought that was sufficient to be able to help me write code for my open source project that I think has a lot of social value. That's why I pay for it. I don't need a buddy or a sexting partner.

I just discovered that I'm paying a monthly subscription to @openaidevs.bsky.social for a coding assistant with enforced dementia. I'm not exaggerating. ChatGPT's short term memory gets dumped randomly every few... minutes? Hours?

Never let it be said that Americans can't figure out a way to turn lemons into lemonade.

Man, if you think we're disrespectful about Charlie Kirk's death, you got no idea how offended you're gonna be when It Finally Happens. 😂😂😂