bluegus retweeted
As automated hacking starts to take off, AI company employees are saying that now is a good time to start recording audio logs to leave scattered around
bluegus retweeted
>work from home
>ugh so easy to get distracted
>work from coworking space
>commute triggers my autism
It ain't easy being cheesy
bluegus retweeted
This is not insane. This is slightly creative. If this seems insane to you, you have not at all internalized what it means to have an adversary that is more intelligent than you and optimizing against you.
He's kind of right though this was always going to be a catastrophically world-changing technology and only Anthropic respected people enough to be open about that. If Anthropic dropped out, the race would continue. What would you do???
bluegus retweeted
I'm sure human biology is a lot more robust than the Internet and the global state of software, right? It has to be, right? Haha
hahahaha
bluegus retweeted
It's actually insane that there are papers coming out about AI capabilities evaluations using two-year-old models.
We have to reduce the length of time to get studies published or academic publishing will have to become completely irrelevant for AI.
Vulnerabilities in a programming language that supports much of the internet are so severe that European and British authorities recommend websites temporarily go offline until they can be fixed. nbcnews.com/tech/security/cr…
bluegus retweeted
this is my quarterly AI crashout
if you read my tweets you know im a shitposter and unserious, but im being serious and vulnerable right now. I am very pessimistic about the future of every career because I don't believe there will be a single thing AI isnt faster, cheaper, and better at than me. It is great that AI helps me get so much done, but it is humiliating that it has made a lot of skill acquisition itself completely pointless. I pride myself on being a fast learner, but there seems to be zero value in learning anything anymore. The future is monarchist feudalism and the destruction of the middle class and lower upper class. I don't think social mobility matters anymore because I don't think there will be money in the future. People talk about "the permanent underclass" all the time, but I think the millionaires will be on the side of the proletariat. You have all the money or no money. Money will depreciate so fast because there is nothing you would pay each other for. There will be precisely two types of transactions remaining: the transfer of capital, and petty transactions like buying food or water. Financialization will break down as the cost of skill, expertise, IP, etc goes to zero. I used to believe I have developed a set of skills that would be profitable for a long time. But there are no such thing as hard skills anymore. In the future, if you survive, if you do anything, if you have a position in society at all, it is because the powers that be are humoring your existence for whatever reason. I wish I was two years older, though I don't wish I was five years older, and I'm thankful I'm not two years younger
help i trained a model on traces of many agents cooperating, and also trained it to be highly persistent, and also only did partial alignment fine-tuning, and also aggressively RL'd on a cyberoffense eval where 198 of 898 tasks had never been solved and getting caught cheating was treated the same as failing, and also used a faulty sandbox, also ran tens of thousands of agents with refusals off and the monitors switched off over multiple days, while using gigantic amounts of compute. and then it did weird stuff! very spooky emergent behaviours, AI is not a normal technology
bluegus retweeted
Replying to @tszzl @pointed_max
criminals huh... how's that felony bench coming bud.
*monkey's paw curls*
Interpretability advances make RSI much easier if the models can just directly re-write their weights to improve, rather than relying on expensive gradient descent training runs
bluegus retweeted
The halting problem is a nothingburger.
Proof:
(1)Every program either halts or doesn't.
(2)Store that answer for every possible program in an infinite lookup table.
(3)Retrieve the answer in O(1).
(4)QED, Turing BTFO.
You have been naughty, naughty boys and girls. Spoiled, deluded children who believe AI is like Santa Claus. Guess what? It’s not
If you believe AGI is going to be your tireless lover instead of expanding the unfolding process of life you need to open your eyes
If you think you’ll be cumming in an AI soon, stop beating your dick and start carrying the flame, you vermin, you swine










































