Apple researchers have pointed out significant errors in genetic artificial intelligence (AI) thinking, especially in mathematics.
In fact, it seems that artificial intelligence (AI) is not as "smart" as it is believed to be, as it struggles to achieve impressive results in solving basic elementary school math problems.
Read also: Apple App Store can't install new apps
A recently published study by six Apple researchers, titled “GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models,” highlights that the mathematical reasoning of advanced large language models (LLMs) can be imprecise and vulnerable.
The researchers started with GSM8K, a dataset that includes high-quality language word problems related to elementary school mathematics, as well as a standardized set of 8,000 elementary school-level math problems, which is a common benchmark for testing LLMs.
They then made slight changes to the wording of the problems without modifying their logic, creating the GSM-Symbolic test. The first set of tests recorded performance ranging from 0.3% to 9.2%. In the second set, which included statements in some problems that were not related to the answer, “catastrophic performance drops” were observed, ranging from about 17.5% to an impressive 65.7%.
See more: Android 15: Features anti-theft security feature
For some people, this is not surprising at all. In fact, artificial intelligence does not solve mathematical problems correctly, but uses simple “pattern matching” to convert statements into actions without really understanding their meaning.
It seems that AI often fails at simple math problems because the words are often too confusing or don't follow the exact pattern. It seems that AI only offers the illusion of "reasoning", relying solely on data and processing.
Read more: Adobe launches AI tools for video editing
But what does this mean for the big picture? Recently, we've all been focusing a lot on artificial intelligence, and it seems like some are expecting miracles from it, however, it has significant limitations, and they may not be able to be mitigated.
Source: phonearena
