I now want to watch an original Trek episode called "How Stands Belshazzar, My Beauteous Love?" in which Kirk destroys AI everywhere in the galaxy by typing the question, "Why?" into it.
Has anyone done a study on how soon the feedback loop of fiction, jokes, and bad info cycling through AI will cause a total breakdown in online information reliability?
The next tech billionaire will be the one to figure out how to filter the AI junk from the internet. And will be, if not bought by Altmann or Musk, richer than both combined.
Not quite that, but roughly 6-7 generations of feedback loops of training on its own output makes it lose coherence (for GPT-3)
More recent algorithms might survive longer, in combination with having more human made data in addition, but it *definitely* isn't good for it to train on its own outputs
I wonder if they can keep AIs from training on their own answers, or if the bad answers get shared to a certain degree if the AIs pick them up from reposts as “valid” and shunt info back into the loop. Obviously I’m not a computer scientist but it’s very interesting.
There's a reason AI companies gave started hiring human writers and fact checkers (... And some of those humans are cheating by passing off questions to another AI)
Gotta love the adorably overwrought nature TOS sometimes had. I believe it still holds the record for Trek series with the most episodes named after Shakespeare lines.
That's become harder now that the primary place people start when looking for information sticks LLM output at the top of the results page, followed by product listings and images, and THEN maybe links to the information you're looking for.
This has been a well known issue for at least 7 months (the "put glue on pizza to keep the cheese from sliding off" story that went viral). If it was as simple as augmenting the prompt or running the query or results through a sentiment filter, this wouldn't still be an issue.
Do they mean the episode "A Fistful of Explosions"? Because part one was badass, but part two was pretty boring when the crew chose Short Circuit for movie night. https://youtu.be/LwDAiAKhsCs?si=le5gydnElR6-ibqc
The AI concludes the opposite of the point of original shitpost. The sentence structure of the original is very simple, and still the AI fails to parse it.
Also it's inventing a whole fictional lore to go with the phrase. "Star Trek fans often jokingly use this to describe a character dying" No they don't! It's a joke from last year but the AI is bullshitting an imaginary context onto it!
This is the high-quality data that Sam Altman is begging to steal for free so he can train the next generation of AI to eliminate human members of Congress and corporate America’s executives.
With AI and robots to do the work and shopping, we can cut back on Homo sapiens, too!
Spoiler alert: In Act IV Kirk gets hit on the head by a coconut and wakes up to discover it was all a dream brought on after Lt. Dr. Helen Noel averted his advances at the Enterprise Christmas party by slipping him a hallucinogen derived from toad venom.
Patrick McGoohan does exactly the same thing in The Prisoner. The only difference is that Kirk has a fistfight in some sand first, while McGoohan is chased by a homicidal weather balloon.
especially since the “references” it digs up are ex post facto justifications for its claims, since an LLM doesn’t “remember” whence it “learned” any particular piece of information
Maybe it's sandbagging. Strategically underperforming on integrity on Star Trek trivia, so that nerdy but imperfect trainers will implement it and allow it to satisfy its goals in the coming Voyager fans vs DS9 fans conflict
JFC, just make the AI find two sources for a fact, & only count news outlets, journals, & encyclopedias as sources, never social media. That would avoid 99% of this nonsense!
Episodes of DS9 are titled "The Midnight Evil That Stirs the Soul" and it's a neo-noir meditation on how some Bajorans resorted to cannibalism during the brutal occupation of their world wrapped in an insightful character piece about one of the side characters.
Comments
Also, if Hypnotoad ever gets a job in the high council, I'll do whatever she asks.
The next tech billionaire will be the one to figure out how to filter the AI junk from the internet. And will be, if not bought by Altmann or Musk, richer than both combined.
More recent algorithms might survive longer, in combination with having more human made data in addition, but it *definitely* isn't good for it to train on its own outputs
Reality: Ctrl+C, Ctrl+V
But also absolutely accurate description of TOS vs TNG
DO NOT GO TO LLMs LIKE ChatGPT FOR INFORMATION.
Do not go to LLMs like ChatGPT for information.
I block anyone still on twitter cause I think you shouldnt be helping a Nazi, but that is just me, a Jewish guy
AI will listen
Careful the things you do
AI will see
and learn
AI is stupid
Careful the things you see
AI will make
You stupid too.
Best title: “For the World Is Hollow and I Have Touched the Sky” - group doesn’t know they’re in an intergenerational
spaceship (hollow world)
"Who Mourns Adonis?"
"For The World Is Hollow and I Have Touched the Sky"
Granted this is third-season heavy but you get the point.
StottAI
nor tell one.
With AI and robots to do the work and shopping, we can cut back on Homo sapiens, too!
"The american dream is to eat as many dead babies as you can stomach, and as soon as you accidentally eat a ceo you frow up" HISTORICAL FACT!!!
All praise the word of Stott 🙌✨
The 18th President of the United States was Hypnotoad.
*makes drinking motion*
Is There In Truth No Beauty
City On the Edge of Forever
(JK. I haven't watched Voyager yet, bf and I started hopping into it the last couple weeks)
*sniff*
...beautiful
It's orders of magnitude funnier because it IS. 💯