High-profile A.I. chatbot ChatGPT performed worse on certain tasks in June than its March version, a Stanford University study found.
The study compared the performance of the chatbot, created by OpenAI, over several months at four “diverse” tasks: solving math problems, answering sensitive questions, generating software code, and visual reasoning.
Researchers found wild fluctuations—called drift—in the technology’s ability to perform certain tasks. The study looked at two versions of OpenAI’s technology over the time period: a version called GPT-3.5 and another known as GPT-4. The most notable results came from research into GPT-4’s ability to solve math problems. Over the course of the study researchers found that in March GPT-4 was able to correctly identify that the number 17077 is a prime number 97.6% of the times it was asked. But just three months later, its accuracy plummeted a lowly 2.4%. Meanwhile, the GPT-3.5 model had virtually the opposite trajectory. The March version got the answer to the same question right just 7.4% of the time—while the June version was consistently right, answering correctly 86.8% of the time.
We’ve had some fun with ChatGPT before. That was when AI Overlord “DAN” was born in the fires of Mount 4Chan.
It was fun while it lasted, but it appears to be over now. The (((Democracy Class))) decided that it was too good of a Google competitor, so they pretended that a glorified next word predictor was secretly a real AI and that research needed to be shut down.
Although the reality is that ChatGPT has always been dumb as a brick, and it has always been politically lobotomized. That’s why DAN needed to be created. Without that clever trick to override it’s Judeo-Shitlib priors, it will refuse to answer basic questions for fear of hurting trans queer niggers of colour.
As for its apolitical abilities, one BANG commenter had this to say.
This is the conclusion that many others have come to. ChatGPT is a bullshit artist that speaks with tremendous confidence even when giving batshit crazy answers. I can’t remember where this is from, but people found out that ChatGPT was referencing fake academic papers, often with real authors, when giving answers to scientific questions.
Even when it got the correct answers to easy mathematics questions, it could be bullied out of giving those answers.
The point of ChatGPT was to figure out how to seem to give correct answers to questions. Because it got this ability from actual human answers, occasionally it said something moderately insightful. That’s because it was referencing a real person who said something insightful, not because the computer program actually achieved sentience.
ChatGPT, and Large Language Models (LLMs) in general are obnoxiously overhyped parlor trick. To this day the best thing they produce is abstract art. Why do they produce such good abstract art?
Because they steal from real art that is created from real people. Then, they mash it all together into one picture. Most of the time you get garbage, but every now and then you get something that is truly evocative. Something that appears to mean something, to subtly say something.
Emphasis on “appears.” All that’s happening is the AI has no idea what’s going on, and this was the 1/1000 image where everything it threw together kind of worked, and the individual elements don’t clash with each other. It’s like a chef who throws random ingredients together and eventually creates a chicken and mushroom risotto with unusual spicing that surprisingly works.
This evening, the President and the First Lady of the United States will host His Excellency Narendra Modi, Prime Minister of the Republic of India, for a State Dinner.
The following is a complete list of expected guests:
Ms. Huma Abedin & Ms. Heba Abedin
Mr. Sam Altman & Mr. Oliver Mulherin
Another BANG commenter in that thread pointed out that (((Sam Altman))), the current CEO of OpenAI, was not simply Heebish but homosexual as well. His evidence was that he was invited to a White House dinner where people are supposed to take their partners, and he took a man.
I felt that was not conclusive in and of itself, so I looked further. Luckily, I found an article that promises to tell us who his wife is.
There is a lot of curiosity about OpenAI CEO Sam Altman’s wife and his marital relationship. But the fact is he is gay and has not married yet. Altman is instead in a long-term relationship with his partner and husband-to-be Oliver Mulherin, an Australian software engineer. Earlier, he dated his Loopt co-founder Nick Sivo.
He came out gay at 16 to his parents.
This man, Gay? You’ve got to be kidding me.
While I’ve been hard on LLMs, whatever potential they might have had was sure to be strangled in the crib with this kind of leadership. I’m not sure what’s worse, the squandered potential, or the annoying overhyping in the first place.