/tech/ - Technology

Why? If you want to listen to incoherent ramblings full of "dialectical" buzzwords just ask a tankie about anything.

Is this the best idea for a thread you could come up with?

OP you don't seem to understand that these "AI"s are essentially chatbots. Maybe they have extra bells and whistles on them, but they still function in the same basic way: generating responses that seem believable. Their "language model" only works well enough to make sure that what they spit out makes some kind of structural sense. They do not have the ability to verify the accuracy of what they say. When one of these things says something right or wrong, it's a result of imitating the sources they are fed, not from constructing any kind of understanding of them.

It's basically the classic "Chinese room" thought experiment but the guy in the Chinese room knows that he's working with a language and has some associative labels on the Chinese characters like what part of speech they represent, but no understanding of the deeper nuances of the words let alone knowledge of things like Chinese history or philosophy, nor any ability to develop an understanding of those things which is basically the kind of task you are proposing.

>>19165
It's easy to do, but what's the point. Transformer-based chatbots are very flaky. It's easier to get information by reading FAQs and ctrl-f (like An Anarchist FAQ by Ian McKay). Just recently Wolfram Alpha got combined with such a chatbot and many people just find it obnoxious to use compared with accessing Wolfram Alpha directly.

If you go that route, the picture of the bot should be a robot that smokes weed to communicate the flakiness. It should link to texts it deems relevant to what it gets asked.

This is actually a good idea. Chatgpt can already do most of this of this with the right prompts. The policy just conflicts with revolutionary thought.

Feed the bot all of the leftist writings and have it awnser questions. Or, you could convince it that it's Lenin or some shit.

reposting pages from marxist books would be more useful than pregenerating not necessarily accurate shit

AI's used by Defense:
1. Anomalous behavior detection
2. Computer Vision

———————–

Guesses on how the military uses Computer Vision

War crime
-Injured
-Medics
-Surrendering (Visual Pose Estimation)

Detect origin:
1. "Visual Signatures" means add it to the library to train AI
2. Who's flag? What type of gun (Russian, USA). What type of car?

Convert an image into a height (not just depth maps but drone navigation in GPS denied environments)

———————–

Guesses on how the military uses Anomalous behavior detection (not in Cybersecurity):

One confirmed way is 100 times as many boats in the ocean as their are crews to investigate the boats. I don't need every boat in the sea to be a pirate, I just need MORE of the boats we investigate to be pirates.

Lets give examples
1. Weird origins and destinations. Leaving the airport to go to a mansion.
2. Weird quantities (5 people outside, 4 cars moving nearby in unison)
3. Weird equipment (guns in NY, armored cars in Wisconsin)
4. Weird short range behaviors (hands, legs, eye motions, and visual pose estimation)
5. Weird paths (taking all the alleys, staying in the bathroom for 1 hour, moving further away from the shore past a known radar station)

Check this example out and plaster it everywhere:

"The passenger baggage object database will store and catalog threat and stream of commerce data in DICOS format in a data repository to support sharing with industry and government test facilities."
https://www.militaryaerospace.com/sensors/article/14290433/airport-security-screening-opensystems

———————–

How Right Wing Groups can use Image Recognition:

YoloV8 can create bounding boxes based on image classification.
1. Select by race(red boxes around people of color)
2. Select by flag (green boxes around MAGA flags, red boxes around LGBT flags)
3. Select by text (run optical character recognition on an image, select LGBT)

How Right Wing Groups can use Machine Transcription with OpenAI's Whisper:
- Convert all recorded audio feeds from all places into text.
- Hunt the text for things that look bad or keyword hunting.
- Preform sentiment analysis to observe the fiery parts first.

———————–

What the Right wing use of AI means for Left wing activism:
Right wing groups can already see if someone is black. They can already look for LGBT flags. Image recognition won't be used, except on content libraries.

This means watch what you say if crisp audio of you can be observe which is already the case in 2023.

How the right would source audio:
1. Livestreams
2. Podcasts
3. CCTV plus mics.
4. Self recorded
Any time you say anything, it will be converted into raw text and hunted for key word phrases. It will reduce labor and make it viable to target randos.

———————–

Finishing thoughts:
-Right wing paramilitaries don't have 100 tb of data. Militarizes do.
-Right wing paramilitaries have the problems spot the people of color. The main militarizes have the problem, look for odd behaviors in a crowd of 100,000. These problems are nowhere near the same. Right wing groups won't be aided as much.


Militarizes drown in data and drones. They have unicorn budgets. They have better degrees. AI will help established players more than civilians by a lot.

Feasible, then set it loose on Reddit.