Oh yes, this AI is just a democratic ploy, right? lol***

quozzel

Hall of Famer [20901]

TigerPulse: 100%

Posts: 12007

Joined: 10/15/02

Re: And the assassination cover-up keeps getting exposed... 'assassination of tr...'

Jul 30, 2024, 4:57 PM

I've been diving into AI's lately because they're incredibly potent tools...how much, non-users just do not understand yet. But anyone not using them - well - is going to be left so far behind they aren't even going to be in the race. Like, for instance, yesterday I wrote a Python back-end app that put a speech-to-text translation model called Vosk into what's known as a "Flask" - for obvious reasons, it's a container for an AI - that allows me to talk to a locally hosted LLM (large language model) AI called "Llama3". On my own computer. Through the use of RAG's - retrieval-augmented generation - I can then plug pretty much anything into it - music players, security systems, spreadsheet programs, public webcam access, internet scrapers - and with my voice alone, tell it to do stuff. The stuff you can do is...amazing. There are guys out there that control fleets of Roomba's, or even give directions to their self-driving cars. From blocks away. With their voice alone. I can already flip between my Ring doorbells just with voice commands.

So basically, I'm sort of a (very) poor man's Tony Stark now, with my very own JARVIS on my desktop. Well, JARVIS's slow-and-special cousin, anyhow. Llama3 is a shockingly good model but it's still very much got some teething problems, like all AI's.

I control it with this python code I wrote yesterday called FlaskedAI.
https://github.com/quozzel/Rando-Stuffz/blob/main/FlaskedAI_v3.py

Sounds impressive, right? It absolutely is. Especially when you figure I'm not actually a developer. I haven't written a line of code since college 20+ years ago, and I don't know Python at all.

AI's do, though. I used ChatGPT-4o to write it for me. So basically, another AI helped make me make my very own JARVIS.

So having dealt with them extensively for more than a year now, I have a very good idea exactly what AI's are capable of, and what they aren't. And they can be very, very smart...and so very, very dumb, and more clueless than any toddler. So, with that in mind, I'll tell you what almost certainly happened there with 99.99% surety.

Basically, you guys happened. AI's are only as good as their inputs, and they do an epic amount of web scraping each and every day. And the Interwebz has gone absolutely bonkers the last few weeks with various conspiracy theories, misinformation, and outright lies. When that happens at sufficient volume, AI's tend to get confused. They've heard so much weirdness and it's been reinforced so many times some of it actually goes through. It's called "noise", and while AI's have noise filters like everything else about AI right now, it's raw and has some bugs and glitches in its code. It definitely was not equipped to deal with The Real Truth About the Second Gunman On the Water Tower.

So, yeah, the AI is clearly broken there. But what broke it was almost certainly not a "conspiracy"...but is actually almost certainly the conspiracy theorists. So congrats, guys. You helped break the Zucker Bot...and are now excited that the Zucker Bot broke. Which is sort of the perfect example of the Conspiracy Theory Circle of Life.

flag link

military_tech

thumb_down

thumb_up

Here’s what google said:

"Our systems have protections against Autocomplete predictions associated with political violence, which were working as intended prior to this horrific event occurring," the spokesperson wrote. "We’re working on improvements to ensure our systems are more up to date."

Since searches with assassination + literally any other name seemed to work just fine, my suspicion is that bias led the filter programming to preclude searches relating to violence and Trump, well before the assassination, thinking Trump supporters were predisposed to violence and trying to head them off at the pass, and the results here were the artifact of that programming.

Could be wrong, but I don’t think it was an attempt to erase Trump so much as it was well-intentioned human bias creeping into the systems. Seems much more likely to me than AI just randomly crapping the bed because it got too many searches.

And the assassination cover-up keeps getting exposed... 'assassination of tr...'

Re: And the assassination cover-up keeps getting exposed... 'assassination of tr...'

Did your own eyes witness J6?***

My eyes witnessed cheating in Charleston Co. Luckily I was there.***

Really??

black people voting***

Re: black people voting***

Are you that obtuse?***

Re: And the assassination cover-up keeps getting exposed... 'assassination of tr...'

Whats being covered up?

Explain what...? Everybody knows what happened. Why try to hide it?***

Even with all the video

Evidence...? It's ALWAYS an "on-going investigation" on a long timeline.***

and we knew this cloud of ambiguity would happen since day one***

How about a dead body.***

Re: And the assassination cover-up keeps getting exposed... 'assassination of tr...'

That was beautiful. Amazing. Astonishing.

Re: That was beautiful. Amazing. Astonishing.

"Assassination of tr???"... That's not by chance or "conspiracy theory" bias.***

who knows?

Its not intentional. Its been fed bad recent data, AI is terrible with

Im hardly pulled into it.

When AI drawings start getting folks fingers correct, call me.

I could be wrong, but I dont think

Here's the first line of ConspiracySolos article.

Hint. I didnt read his article. I dont know what it says.

That bias isnt programmed. It misinterpreted too much incomplete data.***

Disagree

Me.ta AI and autocomplete suggestions are entirely different things***

Absolutely correct, which is my point

RTFA.***

Sure, sure, you literally just said it was intentional.***

Read my reply to IE.

Sure, someone intentionally coded the stack, but there was zero intentionality

TigerBoard POTD