AI’s inner workings remain a mystery, even to its makers

#AIs #workings #remain #mystery #makers

New York: Even the generous artificial intelligence, the technology that expects the world, acknowledges the world, acknowledges that they do not fully understand how these digital systems think.

In an online article published in April, Anthropic co -founder Dario Amody wrote, “People outside the field are often surprised and scared that we do not think how our own AI creations work.”

He added, “This lack of understanding is basically unprecedented in the history of technology.”

Unlike traditional software programs, the programmers learn to go on their way to deliver the results, the Generative AI (General AI) models learn to deliver the results.

In a recent podcast, Chris Olaah, who was part of the open -minded manner before joining the Anthropic, described General AI as “scaffolding” on which circuits increase.

Ola is considered an authority in the so -called mechanical interpretation, to find out how they work, a way of reverse engineering AI models.

This science, which was born almost a decade ago, tries to determine how AI gets a question.

“The fully grip of a big language model is a wonderful act,” said Neil Nanda, a senior research scientist at Google Deep Mind AI Lab.

Nanda added in AFP, “Trying to fully understand the human brain was somewhat resembling”, note that neuro scientists have not yet succeeded on the front.

Taking an interest in digital brains to understand your internal tasks has become a hot area of ​​educational study from a small -known field a few years ago.

“Students are very attracted to this because they feel the effects,” said Mark Crowela, a computer science professor at Boston University.

The professor added that the area of ​​the study is also gaining traction due to the ability to make General AI even more powerful, and because peeking in digital brains can be wisely interesting.

Keeping Ai honest

According to Crowella, the mechanical interpretation not only includes the results presented by General AI, while the technology related to this technology has been calculated while technology raises questions.

The professor explained, “You can consider the model … observe the counts that are being performed and try to understand them.”

Startup Goodfire uses AI software that is capable of representing data in the form of reasoning measures to better understand Gen AI processing and correct errors.

The purpose of this device is also to prevent the General AI model from being maliciously used or deciding to deceive humans, which they deceive humans.

“Goodfire’s chief executive Eric Hoh said,” The most intelligent AI in the world feels like a run against the time to get there before implementing models.

In his article, Amody said that the recent development has hoped that the key to fully understanding the AI ​​will be found in two years.

“I agree that by 2027, we can interpret it, which, reliablely detects model prejudices and harmful intentions,” said Nigivian, an associate professor at the University of Obran University.

According to Boston University’s Coyla, researchers can already access the representation of every digital neuron in the AI ​​brain.

“Unlike the human brain, we have equal to every neuron within these models,” said the educational. “Whatever happens inside the model is fully known for us. This is a question of inquiring the right way to question it.”

Amoodi said that using the internal works of General AI minds can clear its adoption in areas where small mistakes like national security can have dramatic consequences.

For Nanda, human discoveries can also be presented by better understanding what General AI is doing, such as Deep Mind’s chess, Al -Faziro has completely revealed new chess tricks that no great masters have ever thought of.

Properly understood, a general AI model that has reliable stamps will benefit competitively in the market.

Such progress of an American company will also win for the nation in the enmity of its technology with China.

Amoodi wrote, “The powerful AI will create the destiny of humanity.

“We deserve our creations before they change our economy, our lives and our future altogether.”