Two interesting reads today from the New York Times, one recent, the other from December.
Recent one first: AI Is Getting More Powerful, but its Hallucinations Are Getting Worse.
Simply put, newer iterations of AI "reasoning systems" being put out by the major AI players are offering up incorrect information -- which their makers call "hallucinations" more than previous iterations, and they're having a hard time understanding why.
Today’s A.I. bots are based on complex mathematical systems that learn their skills by analyzing enormous amounts of digital data. They do not — and cannot — decide what is true and what is false. Sometimes, they just make stuff up, a phenomenon some A.I. researchers call hallucinations. On one test, the hallucination rates of newer A.I. systems were as high as 79 percent.
These systems use mathematical probabilities to guess the best response, not a strict set of rules defined by human engineers. So they make a certain number of mistakes. “Despite our best efforts, they will always hallucinate,” said Amr Awadallah, the chief executive of Vectara, a start-up that builds A.I. tools for businesses, and a former Google executive. “That will never go away.”
The older article from December of last year offers a few hints as to why.
Is the Tech Industry Already on the Cusp of an A.I. Slowdown?
Interviews with 20 executives and researchers showed a widespread belief that the tech industry is running into a problem that to many was unthinkable just a few years ago: They have used up most of the digital text available on the internet.
They are developing ways for large language models to learn from their own trial and error. By working through various math problems, for instance, language models can learn which methods lead to the right answer and which do not. In essence, the models train on data that they themselves generate. Researchers call this “synthetic data.”
OpenAI recently released a new system, called OpenAI o1, that was built this way. But the method works only in areas like math and computing programming, where there is a firm distinction between right and wrong.
Bold emphasis is mine.
Yikes. They have run out of English text to scrape off the internet. They've fed their large language models everything else. Now they're training on stuff their own AI engines have created. No wonder hallucinations are up.
These engines aren't thinking; they're not intelligent. They can detect and repeat patterns, but they have no idea what is right or wrong. And that's essential to intelligence, in my book.
No comments:
Post a Comment