SCIENTISTS DISCOVER DIGITAL BRAIN TUMORS THAT ACTUALLY IMPROVE AI’S ABILITY TO LIE ABOUT REALITY
In what experts are calling “the best accidental discovery since penicillin and erectile dysfunction pills,” MIT researchers have stumbled upon a way to make AI systems hallucinate more convincing fake images without all that pesky “training” and “effort” previously thought necessary.
LOCAL NERD ACCIDENTALLY SOLVES BILLION-DOLLAR PROBLEM DURING HOMEWORK
Graduate student Lukas Lao Beyer was reportedly just trying to avoid failing a seminar when he discovered that the tiny compressed bits of information inside AI systems could be manipulated like a digital acid trip. What started as a desperate attempt to meet a deadline has now evolved into technology that makes creating fake evidence of your ex-boyfriend riding a unicorn into a volcano easier than ever before.
“I was just messing around, honestly,” admitted Lao Beyer, who will now never need to work again. “I swapped some numbers and suddenly the computer started showing me things that would make Salvador Dalí say ‘that’s a bit much.'”
TOKENS: THE CRYPTOCURRENCY YOU CAN’T INVEST IN BUT WILL SOMEHOW STILL RUIN SOCIETY
The breakthrough revolves around what researchers call “tokens” – not the kind you use at Dave & Buster’s, but tiny mathematical representations that compress entire images into just 32 numbers. Previous systems needed 256 numbers like some kind of digital hoarder.
“It’s like a vocabulary of 4,000 words that makes up an abstract, hidden language spoken by the computer,” explained Professor Kaiming He, apparently unaware of how f@#king terrifying that sentence is to normal humans.
VISUAL HALLUCINATIONS NOW AVAILABLE WITHOUT RECREATIONAL DRUGS
The most jaw-dropping aspect of the research is that these scientists figured out how to generate completely new images without using an image generator, which is sort of like cooking without a stove, or having sex without another person.
“We basically discovered we can just take these tokens, scramble them like eggs, and watch as the computer hallucinates a completely new reality,” said team member Tianhong Li. “It’s like watching a robot have a psychotic break, but in a useful way.”
EXPERTS PREDICT 98% INCREASE IN “NO THAT’S DEFINITELY PHOTOSHOPPED” ARGUMENTS
Dr. Iam Notreel, Professor of Digital Paranoia at the University of Your Worst Fears, warns that this advancement means approximately 87.3% of internet users will soon question whether anything they see is real.
“Before this technology, creating fake images required specialized skills or at least a basic understanding of Photoshop,” said Notreel. “Now your grandmother can make convincing evidence that aliens stole her hydrangeas while she waits for her tea to steep.”
THE ECONOMY OF MAKING SH!T UP IS BOOMING
Industry analysts predict the AI image generation market will reach eleventy billion dollars by next Tuesday, with applications ranging from “legitimate artistic pursuits” to “making your ex look bald on Instagram.”
“We’re particularly excited about applications in self-driving cars,” said Professor Sertac Karaman, apparently unaware that combining hallucinating AI with two-ton metal death machines might raise some concerns.
SILICON VALLEY INVESTORS ALREADY THROWING MONEY AT “WHAT IF WE MADE THE FAKE STUFF EVEN FAKER?”
Venture capitalist Chad Moneybags has reportedly already committed $50 million to a startup called “RealFake,” which promises to “revolutionize the way humans distinguish between reality and computer-generated bullsh!t by making it completely impossible.”
In related news, journalism schools nationwide have updated their curricula to include the course “Is That Actually Happening Or Is It Just A Mathematical Hallucination?: Reporting In The Token Age.”
At press time, researchers were reportedly working on a way to make the technology automatically generate a believable excuse for why you’re late to work that doesn’t involve tokens, tokenizers, or your boss discovering that nothing they see can be trusted ever again.
