Boney Patterson
@bonniepatterson.bsky.social
1K followers 1.1K following 1.8K posts
Lead Narrative Designer on UNANNOUNCED CONSOLE SHOOTER. Yes, we finally get to shoot all those unannounced consoles! She/her. Former cyberneticist.
Posts Media Videos Starter Packs
I don't think it would be that difficult for the Tories. After all, about a third of them were under investigation for "sexual misconduct" - they could catch up with Trump easily.
And they moved Ghislaine to Club Fed to stop her saying that.
That stance suddenly becomes very accessible when you consider that's basically what US insurance companies did with insulin.

"Insulin is based on a public patent."
"No. Our insulin is unique. We made changes."
"What changes?"
"This is strawberry flavoured."

No real difference. Huge cost.
OH! That was it. I just watched an AMAZING mostly found footage K-horror film - they made an incredibly scary badness very simply and effectively and I almost fell out of the window trying not to be in the room with it any more.

It's called "Gonijam: Haunted Asylum".
My parents had a photo of me doing exactly that.

With a snake.

Oh, same year too!
Shit, I came on here to post something, and I forgot what it was.
The "when you are sick" thing is so bloody aggravating.

"OK, so you want to start wearing a mask after you've already infected 3-9 people."
1. "Dear MMO Player, plz find enclosed your Shield of Freezing +1"
2. The worst possible way to see what's on the other side of your negatives.
3. Duck translator
4. "Dear MMO Player, your Shield of Freezing +1 has frozen. Plz stand by."
Well, now I'll *never* get this Sudoku puzzle finished.
Yet governments have AI symposiums and devote most of the time in them to "Will robots destroy the world?"

(No, they won't, they'll destroy an empty volume of space 300 AU away, and by "destroy", I mean "put cheese in it".)

But will they convince enough investors to replace staff with AI? Yep
...self-driving demo was not self-driving. It was staff at Tesla HQ remote-driving the car. A man in a robot suit dancing. People in third-world countries Mechanical Turking to type in answers to questions.
And you're absolutely right: I don't understand why we are being forced to use this inferior version of a search engine, replace talented artists with tentacle hands, lay off thousands of people, to do a worse job more slowly with AI.

They are literally faking their demos. Tesla's...
(I love your username)

It's the gold rush all over again. Desperate people set out to look for gold, and oh no! They have no picks or shovels to dig with. "We'll sell you picks and shovels" say the AI dealers of the world and start pocketing the cash.
If anything is worth taking away from my rantings, it's this: Taking something that is meant to learn and generate its own responses without using a fixed table of answers, and stapling it to a fixed table of answers, is exactly how you design a problem.
⬆️partway done, delivering very lurid bollocks.

Though plz note:

1) It's been a long time since I dealt with this stuff
2) I no longer have the resources to model and test

So take with a large pinch of salt anything I say
⬆️it tries to adjust its weights and thresholds to "learn". But the LLM is dumber than a refrigerator and cannot learn, so it sounds like they're looping. A loop is very expensive and annoying to users so they probably have a forced exit and that's your hallucination, with the NN adjustments only⬇️
⬆️static probability table of words, colour codes, equations etc etc trained on the smallest and most F**KING DUMBASS portion of online knowledge (social media)? The AI comes to its answers (albeit somewhat constrained by its size), and if it doesn't land in an area the LLM deems acceptable, ⬇️
⬆️have an extraordinary number of neurons (only 3 layers deep, but very wide) to achieve the kind of generality we are being sold. But how are LLM's the right way to solve that problem? Why take a possibly-intelligent system that can generate a wide variety of suitable responses and nail it to a ⬇️
Seriously, folks, caylen is not wrong. LLM-based AI *does not work*. Think about it: you have neural networks. They take training data in and have their weight and threshold values adjusted incrementally until they start producing the right output. They don't need an LLM. But they would need to...⬇️
Reposted by Boney Patterson
you fucked around and now the Episcopalians are doing memes. are you happy now. are you
Clive Barker would have loved these in his Cabal era (The past tense is not casting aspersions, there, I just don't know what he's into now that he's happy)