Progress is rarely written, and AI is not an exception.
As academics, independent developers, and the largest technology companies in the world, they approach us from artificial general intelligence – a virtual form of intelligence that still corresponds to human capabilities – they hit some road barriers. Many emerging models are vulnerable to hallucinations, wrong information and simple mistakes.
Google Sundar Pichai CEO of Google Pichai referred to this stage of AI in the name of AJI, or “Artificial Intelligence”, in a recent episode of Lex Fridman.
In reference to the deep learning and computer vision specialist Andrig Carbashi, who founded Openai before leaving it last year: “I don’t know who I used first, maybe Karpathy did,” he said, referring to a deep learning and computer vision specialist, who founded Openai before leaving last year, “said Bishi, referring to a deep learning and computer vision specialist Andrig Carbashi, who founded Openai before leaving last year.
AJI is part of a metaphor for the path of artificial intelligence development – rough, and simultaneously occurs sparks of genius and basic errors.
In the publication of 2024 x entitled “Wooden Intelligence”, Karpathy described the term as “the word it reached to describe the truth (strange, non -intuitive) that the LLMS case can perform very impressive tasks (such as solving complex mathematics problems) at the same time with some very stupid problems.” Then he published examples of the latest artistic language models that fail to understand that 9.9 is greater than 9.11, making “unusual decisions” in Tic-Toe game, and fight for dependence.
Carbashi said that the issue is that unlike humans, “when many knowledge capabilities and problem solving are largely related and all linear improvement together, from birth to adulthood,” said Karpathy.
Peshay repeated the idea.
“You see what they can do, and after that you can find trivial that they are making digital errors, Red in strawberries or something, which seems to wander around most models,” said Bishi. “I feel that we are in Aji where dramatic progress, some things do not work well, but in general, you see a lot of progress.”
In 2010, when Google DeepMind was launched, her team will talk about a 20 -year schedule for AGI, Bishy said. Google then got DeepMind in 2014. Pichai believes that it will take a little longer, but by 2030, “I would like to confirm that this definition does not matter because you will have progress in thinking about many dimensions.”
By that time, the world will also need a clear system to describe the content created by artificial intelligence “to distinguish reality”.
“Progress” is a mysterious term, but Pachai spoke at length about the benefits that we will see from the development of artificial intelligence. At the United Nations Future Summit in September 2024, four specific methods have been identified by Amnesty International – to improve access to knowledge of original languages, accelerate scientific discovery, reduce the climate disaster, and contribute to economic progress.
However, first, he needs to do so Learn to spell “strawberry.”
