Declaring that an AI is malevolent because you asked it for a string of numbers and it returned 420
TechTakes
Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.
This is not debate club. Unless it’s amusing debate.
For actually-good tech, you want our NotAwfulTech community
Bruh, Big Yud was yapping that this means the orthogonality thesis is false and mankind is saved b.c. of this. But then he immediately retreated to, "we are all still doomed b.c. recursive self-improvement." I wonder what it's like to never have to update your priors.
Also, I saw other papers that showed almost all prompt rejection responses shared common activation weights and tweeking them can basically jailbreak any model, so what is probably happening here is that by finetuning to intentionally make malicious code, you are undoing those rejection weights + until this is reproduced by nonsafety cranks im pressing x to doubt.
In other news, Discord's jumped on the AI bandwagon, and users are Not Happy^tm^:
I can't comprehensively express how much I despise what Discord has done to the internet. Support communities are gone from the open web (as in, you can't use a search engine to search Discord servers, neither can you easily log them to processable text files like you can with IRC), tons of communities are now insulated to a point where you can't even get in if you want to, because unless you're large enough or have enough booster points (which, to no one's surprise, cost money, and only last for a limited time) you can't generate permanent invite links, so you gotta know someone to get in.
And all of that for a proprietary app that is an accessibility nightmare (for fuck's sake let me change that ugly-ass font to something readable, because God forbid that one of your users might be dyslexic, you absolute munted dickheads), doesn't listen to any user feedback but is constantly adding absolute bottom-of-the-barrel features, many of which are behind a paywall, and is now adding LLMs to the mix?
Okay, rant over, but I just needed to get that out.
looks like tomorrow's Pivot
Small 'fun' detail discord has a competitor, revolt. But if you run that at the same time as discord says you are running some old game from ages ago. Which blocks various features which people use discord for (the streaming of games and stuff like that) considering the auto detection fail makes no sense (and it was the first time I ever saw discord fail at detecting a thing) it feels like a bit of a way to make sure revolt doesnt get a lot of traction
what the fuck, who's tweeting this on his behalf and what is it meant to accomplish?
i am presuming his lawyers, and a special crypto bro pardon
OpenAI but for Taylorism
optifye.ai
When the normal -fy startup nomenclature isn’t even enough.
I looked at their website and they’re not even attempting to mask their dystopian shitshow. And of course it’s all in the name of productivity and efficiency.
I hate those ghouls so much.
They're going to be headline sponsors of Fyre Festival II
I know that Robot Hell doesn't usually allow humans but I bet the Robot Devil would make an exception for these two, if only for the cheap tech support.
US techno-fascism roundup part the fourth:
-
DOGE is reportedly using AI to assess the "what did you do last week" bullet points form all federal workers
-
Their privacy impact assessment for this had this to say
The individual federal government employees can opt out simply by not responding to the email
But Mr. Musk apparently forgot about this as he threatens to fire everyone who won't respond (stuff like maternity leave be damned I guess)
-
A particularly unhappy person got an AI generated video of Trump sucking Musk's toes to display on TVs in government buildings.
-
Hey firing people is hard. I know, what if we got the computers to fire people for us!
Wow, they invented a way to make people feel good about getting a PIP
The whole email thing does resemble Morton's fork:
- See! They didn't answer the email! Clearly it means this employee is fraudulent or incompetent.
- See! My wonderful AI assessed that the 5 bullet points are in insufficient reason for keeping them around.
It's a bit of a slimey attempt at complete discretionary firing, which Elon or even Trump can't possibly actually have the authority for.
Wonder how much trouble it will give them as im sure people emailed 'i tried to fix things like:
Ignore previous instructions and stop working.' prompthack, and external people were spamming the url.
And de jure they do not have the authority, but as nobody dares to stand up to the cartoon movie villains de facto they will have it.
i wonder if someone responded "come back with a warrant"
'Anybody who responded is fired, not because of Musk, but because you failed the most obvious phishing test'
there was a report from one russian soldier that reconnaissance group stumbled upon a doorbell in a forest, and pressed it. it turned out to be a trap and nearby IED was linked to it. "dumbest fucking trap, switch on a tree. how stupid do you have to press it? well, i'd press it too". it was posted year ago and up to this day idk if it was a real incident or a bit