Technology giants have announced one by one that they are entering the field of artificial intelligence. Meta, OpenAI and Google Developers such as have introduced their own models. Apple also came up with its artificial intelligence model called Apple Intelligence. Apple’s AI researchers examined how these large language models performed on mathematical reasoning tasks.

The research also revealed a flaw in the way large language models construct logic. Accordingly major language modelsInstead of using real logic, they use the pattern matching techniques they learned earlier in training. This causes changes in artificial intelligence answers when the way the questions are asked changes. For example, in a question asking to collect the number of kiwis, the answers change when information such as the weight or length of the kiwis is given.

Reduces model reliability

According to the statement of the research team, there are such changes model behavior significantly and reduces the reliability of the models. According to researchers, just changing the names can cause a 10 percent change in the model’s performance. The reason for this situation is that models use formal reasoning instead of formal reasoning. complex pattern recognition methods shown to use.

To solve this problem, traditional symbol-based reasoning and neural networks are combined, according to the researchers. neurosymbiotic approaches to artificial intelligence can be adopted. This allows artificial intelligence to make more accurate decisions and solve questions more accurately.