Listen to this story
|
In his lecture at Oxford University, Geoffrey Hinton, one of three godfathers of AI, said that digital models are already as close to being as good as brains and will eventually get better than brains.
“A large language model has a trillion weights. You have 100 trillion weights. Even if you use 10% of that, you have 10 trillion weights,” said Hinton. He adds that an LLM in its trillion weights knows thousands of times more than we do. “It’s got much more knowledge and that’s partly because it has seen much more data,” he explains, that it might also be because it has much better learning algorithms.
He explains that humans are not optimised for packaging lots of experience into connections. “We are optimised for not having many experiences. You only live for about a billion seconds,” Hinton adds, saying that humans don’t really learn anything after the age of 30. “You have got crazily more parameters than you have got experiences. Our brain is optimised for not having many experiences.”
Recently, Elon Musk also reshared a video from the Joe Rogan podcast with Ray Kurzweil that AI will match the knowledge of any person by 2029. “AI will probably be smarter than any single human next year. By 2029, AI is probably smarter than all humans combined,” said Musk.
Kurzweil explained that people often underestimate the rate at which technology grows. “It actually doubles in fourteen years,” whereas, he said, people think it only grows by 2% every year. Talking about the speed of computers, he said that the calculation speed has increased to 35 billion calculations per second, which is a 24 quadrillion fold increase from 0.00007 calculations per second in 1939.
Hinton, who left Google last year, also has been openly also discussing the threats of AI. He compares the potential risks of AI to the creation of the atomic bomb during World War II, emphasising the dangers of profit-driven AI development that could result in AI-generated content surpassing human-produced content and jeopardising our survival.