Meta and Apple boldly state: The emperor of genAI stands naked
“According to him, the current algorithms are simply anticipating the following word in a text. They are incredibly proficient at this, deceiving us in the process.
“According to him, the current algorithms are simply anticipating the following word in a text. They are incredibly proficient at this, deceiving us in the process. Their vast memory capacity makes them appear to be reasoning, when in reality they are just repeating information they have learned before.
“‘We are accustomed to associating intelligence with the ability to express oneself or manipulate language, but that’s a misconception,’ remarked LeCun. ‘You can manipulate language without being intelligent, and that’s essentially what LLMs are revealing.’”
This presents a fundamental challenge. Many businesses are placing excessive trust in genAI systems, asserted Francesco Perticarari, a principal partner at the tech investment firm Silicon Roundabout Ventures based in London, England.
It is common to interpret the occasional correct responses from these tools as strokes of brilliance, rather than the genAI making fortunate guesses. However, “the outcomes are not driven by reasoning, but solely by powerful computational capabilities,” highlighted Perticarari.
Empowering genAI
One popular argument in favor of genAI is that some models have demonstrated effectiveness in passing different state bar exams. Yet, these exams provide an ideal setting for genAI as all answers are already known. GenAI excels in memorization and repetition, but that does not indicate that genAI possesses the legal skills, comprehension, and judgement required for practicing law.
About Author
