Printable View
so what do you think happened last night? i was up late and twitter was afire. then apparently when it hit the wrong/right nerve - jews- shit got shut down.
pls enlighten us neophytes with your two-cents, as well as thoughts on what it means for such things moving forward.
i think we tucked into this in the elon thread but last week elon said he was personally taking over grok's training because it was too woke.
fast forward to today, elon lost a director of engineering and a ceo in like 12 hours after his bot went full /pol/, sexually harassed linda, and then described raping will stancil for good measure.
i dont know, i dont have much to add really. something jim simons said a long time ago proves itself more and more true with each passing day; if youre developing a system that makes deterministic judgements based on data, the vast majority of your time is going to be spent cleaning that data. its true with alg trading and its true with chatbots and true with anything else that requires developing a model.
and something they get into with real glorious detail here https://www.systemicmisalignment.com is that once you train your model, all the personalities you hoovered up are in it for good. you can filter out the monster but its never going away entirely and that process of tweaking and filtering is incredibly esoteric, hence the scifi level salaries it pays.
but elon always thinks he knows better and that the secret to life is to never give a shit about anything but himself because none of it matters and no one else has real feelings but him.
and here we are.
Quote:
Originally Posted by sonatine
so, entropy w/ AI?
whats the old expression.. garbage in, garbage out.
When a large language model responds to an abstract questionQuote:
Originally Posted by Tellafriend
by endorsing the most hated person of our times you have to first
follow-up with the question of why? Bots don't understand the why question. An AI response to the second question would likely be nonsensical
Whether it's an AI bot or an anonymous contributor on a social platform all must be viewed as suspect.
The ability to disseminate misinformation, disinformation and hate based speech anonymously is a reality. This weapon is largely unchecked because naive or ignorant people believe it to be a right
Anonymity is the key. People and or groups have an avenue to express their most hate filled thoughts without consequence.
Pedophiles beware. On-line anonymity is not a lock.
As with many aspects of American society, the gloves (regulations) are off. You can buy that AR-15 along with a bottle of Jack at the next off-ramp because you are free. You are also free to vote for
1 of 2 corporation owned, lobbyist influenced political parties
AI w no white people- hahaha
AI said sometyhing Jews don't like- This is serious, cue congressional subpeonas. Turn over algorythym to the ADL. CEO resigns, banktupt the company, lawsuits galore It's still day 1
we are so back
Strangely enough, I knew Nikita Bier's family when he was a little kid.
He and his brother were running wild with little supervision, and always getting into trouble.
I figured that in 30 or so years, the updates about his life wouldn't be good. Instead, the dude has made big money by developing apps which appear to threaten the dominance of some major company's brand, gets bought out, and pockets huge money.
Interesting that he's working at X now.
Really bright (albeit very cynical) guy, but funny thinking back to him as a wild little kid. At least when I knew 10-year-old Joseph Gordon-Levitt, I had a feeling he was going somewhere.
Anyway, Grok's worship of Hitler isn't a surprise.
These things aren't human. They learn from what they read. AI bots are constantly trying to scan PFA all day and night, which is why we had so many server slowdowns over the past 2 years or so. Finally I had to do mass bans of any kind of semi-suspicious traffic in order to thwart it.
It read a lot of shit about how the Jews are evil, Hitler was an unfairly vilified good leader, etc, and vomits it back out.
Keep in mind that a lot of Grok's sources are X itself. I know this from asking it about me.
https://i.imgur.com/qLeqZHd.jpeg
not a word of this is true.
grok 4 looks vaguely familiar
https://x.com/TonyMichaelX/status/1943956670255702423
To date AI is obvious.
Why do scammers employing AI allow errors in the finished product?
AI has trouble with numbers, weights, dates or local pronunciations
March 2nd, an abbreviation, is not recognized or repeated properly by today's AI.
Advertisers don't seem to care. They can't be bothered to make corrections.
Being skeptical of advertisements is rule #1. Any error in the pitch
is a red flag. When a "trusted" or known person appears to be endorsing snake oil, there will be casualties.
https://pbs.twimg.com/media/GvyUi9KX...pg&name=medium
so this failorgy kinda compliments what i was saying above.
they trained on data that included perspectives they didnt want, which elon considered 'woke'.
elon, feeling left out of the AI relevance conversation, declares that he alone could 'fix' grok and about 3 days later, they push something new and exciting to prod. now admittedly we dont know exactly what happened; we dont know if it was a new model, or if they applied weights to skew it towards elons world view, but the next 16 hours will forever be known as the mechahitler era.
and then they apparently went back to the original model but again attempted to stifle the personalities that elon disproved of, this time by creating a logic gate that took every outbound opinion and only ran with it if it was an opinion elon shared. if it conflicted with elons world view, elons world view was selected.
which of course crashed and burned.
and now we are in the grok 4 phase where elon, having burned through a few billion dollars and shed a significant percent of his executive staff, now admits that he has to create a brand new model from the ground up, ostensibly trained off stormfront and /pol/ but with some currently non-existent anti-extremism gate, which wont work and will again end up routed through just_say_what_elon_think(). which will likely cost him more executives and billions of dollars.
to which he responds by telling his investors that hes effectively grabbing $200bn of tesla's capital reserves to 'invest' in xAI, which will finance training grok and all that comes with that, but dont worry because being the guy with his finger on the pulse of the average american, he has realized that the thing his cars need more than mileage or innovation or reliability is an ultra right wing white nationalist chatbot, and this will be the masterstroke that defeats the woke mind virus.
but yeah the only thing thats going to solve this is him dropping acid and listening to radiohead and having a spiritual awakening and until then, hes just going to burn through every form of capital at his disposal to win back grimes and/or humiliate his enemies.