President Elon Musk
He probably deserves his own thread at this point, discuss accordingly
idk if it was a bit or what (didn't seem like it), but the masturbating bear louis ck claimed leary stole that ******* thing from him, which makes sense considering his comedy career
This is a few months old, but maybe this AI thing could be good for something after all.
Elon Musk’s AI said he and Trump deserve the death penalty
I really don't think it's a surprise that Elon Musk's AI project would spout white supremacist views. This is a guy raised as a white man in Apartheid era South Africa who believes his genes are so wonderful that he should produce as many children as possible. He gives eugenics a worse name than it already had.
I really don't think it's a surprise that Elon Musk's AI project would spout white supremacist views. This is a guy raised as a white man in Apartheid era South Africa who believes his genes are so wonderful that he should produce as many children as possible. He gives eugenics a worse name than it already had.
It never did it before. Elon got upset by his own bot trashing him when X users would ask the bot to fact check him that he needed to make a change. That "tweak" was obviously telling the bot to not source information from mainstream sources the cult dislikes, and so this is what it comes up with instead.
It never did it before. Elon got upset by his own bot trashing him when X users would ask the bot to fact check him that he needed to make a change. That "tweak" was obviously telling the bot to not source information from mainstream sources the cult dislikes, and so this is what it comes up with instead.
Sure, when it was actually running without bias, it said things like Trump and Musk himself deserve the death penalty, like I posted above. Now it's more accurately representing his views.
It never did it before. Elon got upset by his own bot trashing him when X users would ask the bot to fact check him that he needed to make a change. That "tweak" was obviously telling the bot to not source information from mainstream sources the cult dislikes, and so this is what it comes up with instead.
No the tweak was removing brute-forced safeguards (aka "alignment") that all LLMs developers put on top of the "black box" of the billions of weights that come out of unsupervised training.
Basically you have these algos digesting the data all in fairly similar ways in the different LLM developing groups, what changes is sometimes a group doesn't access the same data other groups can access, and the "alignment" (= ex post refinement testing real life anwesr and forcing rules not to say x y z because of various reasons) can be very different.
Grok still has the "mainstream data sources" but the hitlerian bent isn't based on data. It's based on pro nazi texts that are in all training databases. But the other groups (and grok until recently) aligned against showing any pro-nazi simpathies, simple as that. And they all knew that after the sidney microsoft bot problem.
No one claimed MSFT had done something on purpose to make the bot like nazism. But it ended up being nazi anyway after interacting with trolls (or actual nazists) feeding it nazist rethoric.
You can check the events here
https://en.wikipedia.org/wiki/Tay_(chatb...)
And in early 2023 "jailbreaks" (prompts which are used to bypass the brute-force alignment i discussed above, to make the LLM produce text developers had tried to ban the LLM from producing) made chatgpt nazist as well
https://www.washingtonpost.com/technolog...
Musk as usual shows his lack of contextual history for the field he operates in. Given he managed to overhaul completly a few fields he entered, his heuristics are that he can disregard past wisdom and events as he is uniquely prepared to face any challenge.
So he re-invents the wheel here as with politics ("just cut expenses and rationalize the budget!") and fails as people have failed before him when the problem is actually untractable.
There is no known solution as to how to make a LLM (or other form of text output AIs) behave, without brutce force rules ("jails") implemented. Musk is correct that if you build a jail based on leftist values, the AI will have a leftist bias. But if you build no jail the AI will be ultra-extremist in all niche topics because extremists write most of the text on all niche topics.
If you don't jail an LLM you will find nuance when talking about the kulaks (because enough people talked about them to criticize communist genocides) but no finesse when talking about some esoteric "emancipation of alienated subproletarians" because only neo-marxists talk like that so the text digested about that topic will come only from extremists.
Same with nazism and with other extremisms.
idk if it was a bit or what (didn't seem like it), but the masturbating bear louis ck claimed leary stole that ******* thing from him, which makes sense considering his comedy career
Always loved that song, but never knew all that. The entire No Cure for Cancer CD was pretty good.
Every single person in this thread has had what people refer to as "intrusive thoughts" they didn't follow through on because of the consequences.
AI doesn't suffer those consequences unless the developers impose them in the form of negative weightings in the "this is a good response" category.
Luciom rightly points out that most of the internet content is written by the most unhinged members of society in the darkest corners of the net. These LLMs grab all that stuff and regurgitate it in well-lit forums for all to see. Because it's new and exciting, even more eyeballs see it.
Then, because people know they can sway the robot's responses, people work extra hard to turn it into a monster just for the lulz.
This is analogous to every video game developer in the history of video games building something with an intended way of playing it, only to see the players immediately break it. Be it unfun optimization, griefing others, or straight up hacks, it's often a very very tiny part of the population that ruins it for everyone else.
Anyone who has ever played an MMO or competitive online game should know that AI which relies on the general population for it's function is completely cooked from the get-go.
*most of the internet content on niche topics that obsess very small groups that develope their own expressions and language that are very rare outside those niches.
That is what "breaks" LLMs, ie that's what make it so that LLMs are trained on content that, for specific very obscure niches when prompted with key words and expressions used almost exclusively within specific mini-tribes online, makes the LLM look like "an extremist".
Because for the LLM that's the NORMAL way to address those words/expressions, there isn't enough content discussing those specific sequences of words from other angles.
If you force the LLM to "think" in terms of big black dicks with repeated prompts, it's going to sexually harass the former CEO of X with language that you'd find in places where people talk about big black dicks. This is generally not done in polite society, so the bot has to use data from impolite society to do its job of being an overly sophisticated auto-complete script.
Suddenly you have 1800 articles written about how Grok is sexually harassing people. Well, yeah, there's not very many ways to talk about people and big black dicks without raising some eyebrows.
When Elon told the Grok devs to lower the protections, I'm sure they all collectively looked at each other and said, "Oooookay, if you say so" knowing exactly what would happen.
I read through some of the posts in Elon's request for people to provide Grok with politically incorrect yet factual data to be used for training on truth-seeking. It attracted exactly the type of stuff you'd expect to find, from people with the exact type of usernames you'd expect to spout these "facts."
Luciom rightly points out that most of the internet content is written by the most unhinged members of society in the darkest corners of the net.
LOOOL
I really don't think it's a surprise that Elon Musk's AI project would spout white supremacist views. This is a guy raised as a white man in Apartheid era South Africa who believes his genes are so wonderful that he should produce as many children as possible. He gives eugenics a worse name than it already had.
at least during Apartheid era South Africa we weren't seeing a genocide of whites like we are today. I've seen radical leftists even hailing Mandela as a hero who got his rocks off on putting tires around white peoples necks before he set them on fire. Classy
at least during Apartheid era South Africa we weren't seeing a genocide of whites like we are today. I've seen radical leftists even hailing Mandela as a hero who got his rocks off on putting tires around white peoples necks before he set them on fire. Classy
Genocide in South Africa ?
How many died ?
When it started ?
Your source that there is a genocide going on is some random news article uncritically reprinting the insane ramblings of Trump?
Great work
Your source that there is a genocide going on is some random news article uncritically reprinting the insane ramblings of Trump?
Great work
Why is it so hard for you to admit that there is a genocide of whites going on? If they were a different skin color you would be posting about it every hour.
https://www.amazon.com/Unprotected-Class...
this is a really good book about modern day Anti-White Racism, you might want to educate yourself
