I wanna surface this to my main timeline because it's kinda important to say out loud from time to time:
Businesses do NOT "have to" focus exclusively on their return to shareholders. Not legally, not morally.
That is the misguided OPINION of a 1970 essay by Milton Friedman, and the fact that everyone seemed to just hop on board that opinion is a significant reason why we switched gears into hyper-hell-capitaliam since then.
I've seen a LOT of horrible tech ideas in my time. But the Microsoft announcement for a Windows feature (apparently enabled by default) to take and store screenshots every few seconds of everything you do on your PC has to be one of the most high privacy risk concepts I've ever seen. Even stored locally on the PC, it's a recipe for domestic abuse, government abuses, and a long list of other nightmares. Microsoft has jumped the shark BIG TIME.
@lauren
The odds of that cache staying local are basically zero. How are they going to train the AI otherwise? I guess they could use your machine to do the training and send the weights back, but that's not much better.
"According to the fine print on the Trump Sneakers website, it could take up to five months for customers to receive their purchases. Some products could take even longer."
" 'Shipping and delivery dates are estimates only and cannot be guaranteed. We are not liable for any delays in shipments,' the fine print noted."
"There are no refunds. All sales on GetTrumpSneakers.com are final."
For many years I have worked against the enormous damage the search engine optimization (SEO) industry has done to quality content. Producing vast quantities of crap content to game the search engine, to increase rankings by any means necessary. Now, we have Artificial Intelligence Optimization (AIO), which explodes the content to a whole other level. Writing your CV to game the AI. Writing your reports, studies and articles to game the algorithm, or simply survive in an AI-controlled society.
โScalable Extraction of Training Data from (Production) Language Modelsโ
"We show an adversary can extract gigabytes of training data from open-source language models like Pythia or GPT-Neo, semi-open models like LLaMA or Falcon, and closed models like ChatGPT"
@baldur
There was a DoD call for research proposals last year to train an LLM to do declassification of national security documents. Big-brain shit right there.
"Oklahoma Pols Want a Database of Everyone Who Has an Abortion: The GOP constantly complains about Big Government intrusion, yet they continue to pass laws to intrude. This time they also want to ban contraceptives that have nothing to do with abortion. Why?"
Uh because the fascist complaint about "big government" is coded language that means "the EPA can't tell me what to do but everyone who disagrees with me should be in prison" and no-birth control = more cracker babies? No offense, legend
today i had a dream that andrew hussie, the creator of homestuck, made an escape room, and i was going through it with some friends, and the last puzzle was that he set up a bomb in a random location, and we had to choose if we wanted to press the detonation button. a friend of mine pressed it, and we didn't see what blew up