I had a long sneer typed out but my shitty Lemmy app deleted my draft. Fuck the orange site and it's sophists. The whole prison subthread is probably one of the most disgusting things I've read from that shit hole in recent memory.
sinedpick
joined 2 years ago
ow fuck! my toe!
What happened with SBF will happen with an AI given a similar target, in terms of having misalignments that start out tolerable but steadily grow worse as capabilities increase and you face situations outside of the distribution, and things start to spiral to places very far than anything you ever would have intended.
Ah yes, one day someone will accidentally install the "I'm sorry, I can't let you do that Hal" plugin. Oops, I let the nuke launch AI override all of our control mechanisms, silly me!
I fucking hate x-risk people so much.
I'm a nixos fanatic, thank you very much! I even installed nixos on my (non-technical) SO's computer. Was this a mistake? Yes. Do I care? no.
orange site: great! I loved thiel but now that I can see him licking boots I love him even more!
Random musings but I feel like posting: I almost got sucked into the urbit hole. I thought it was such a cool idea and implementation, and all the fun names and technical purity was so attractive to me at age 18 (this was before the crypto circus and urbit barely did anything except talk to a terminal).
It took me a while to realize that there is, actually, zero reason to give nonsensical names to literally every aspect of software, and also pretty dumb to try to shoehorn everything through a tiny functional core ("Nock") while slowly re-learning all the lessons of 50 years of compiler development. So why use it at all?
Using urbit over a normal Linux stack comes purely with downsides. Slow, buggy, obscurantist, and so on. This means whoever actually dedicates their precious time to developing this unconditionally buys into the ideology. I never thought an ideology could be so powerful that it could corrupt the minds of my people (software monkeys).
Urbit is a truly fascist^H^Hnating phenomenon.
A reminder that Rationalists have absolutely No Fucking Clue what they're talking about when it comes to quantum mechanics, and this is evident from the very top.
Here is their prophet's, Eliezer Yudkowsky's, brilliant writings on QM: https://www.lesswrong.com/posts/5vZD32EynD9n94dhr/configurations-and-amplitude
In this stunning vindication of Dunning-Kruger, EY sets up a thought experiment of a photon being ejected at a half-silvered mirror. Then, he realizes that QM is formulated with complex numbers, so he decides to shoehorn them by imagining a "computer program" that computes the result of the experiment and using the complex numbers as the internal state (because he read somewhere that a wave function is a complex-valued function). From there, he goes on to realize that he needs to actually justify the use of complex numbers, so he drops the fact that multiplying the "internal state" by i represents the photon turning 90 degrees (what?! yes, multiplying by i rotates complex numbers by 90 degrees but this has literally nothing to do with the direction the photon travels, what the ACTUAL fuck am I reading?)
I seriously want to pull my hair out after reading this asinine nonsense. MIT OCW's QM course is extremely accessible to anyone with a decent high-school math education but these chucklefucks' need to prove to themselves that they're smart supercedes any process of actual learning.
edit because I can't stop sneering: "wave function collapse" is purely born of the Copenhagen interpretation which EY rails against as ridiculous (which, admittedly, isn't a totally unpopular opinion for real physicists to have). This is, of course, 100% lost on SBF.
It would be extremely convenient for Sam "16 million dollar penthouse" Bankman-Fried if this whole deal was just an "oopsie, silly me." I get the schadenfreude angle but this guy totally doesn't deserve freedom.
What I'm the most thrilled about, however, is how he has effectively dismantled the rational case for "effective altruism" by showing how mind-bogglingly stupid the idea "earn to give" is. It's so easy to shut them down now with 3 magic letters: SBF
The amount of lazy "it is what it is" takes makes me want to vomit.
Every system has some form of bias, more or less, and a system that has less of a functional bias than another system isn't necessarily a better one
I can't even begin to comprehend how asinine this take is.
Nah bro, when GPT-5 comes out all code it'll write will exactly match the specification, and it'll also sim the entire universe to guess your mental state and correct any mistakes you made in your specs.
You were asked to give a use-case for LLMs, and with this comes the implicit assumption that it's not something that can be easily done with a tool that costs about seven orders of magnitude less to produce and operate.
A bunch of junior devs writing repetitive code because it's easier or people refusing to learn proper tools because "AI can write my JSON" aren't exactly good reasons tor the rest of the industry to learn how LLMs work. Don't get me wrong, there are good reasons, but you've not listed any.
For how much these fucks talk about "overcoming bias", they seem ABSOLUTELY incapable of overcoming their bias for IQ as a good measure of anything outside of acute mental disability. They want a simple answer to the question that has plagued their small minds forever: "Am I smarter than that person over there?" They cling to their number like a life-preserver in the ocean of society.
It's just so pathetic.
This is hard for me to digest because I have a lot of respect for Karpathy. His famous article on the potential of LSTMs as language models absolutely blew my mind. The innovations that led to LLMs stand on his shoulders, so how can he have such a poor take here?