According to a new study, which was published by the research team of the Apple, the larger IT shows the weakness of experience in reasoning and logic. The study examines the ability of the various models of language, including those developed by OpenAI, and the Spot to handle the tasks of reasoning mathematical. What stands out is that even small changes in the wording of the questions may have an impact on the responses.
The Apple: a study on the limits and he he he he HE
Apple points to the fact that the models of artificial intelligence tends to be based on the recognition of patterns, and not in the reasoning and logic of the current. This is a factor that has the potential to be detrimental, provided that the changes are minor in the context, or the information can lead to the reaction significantly different.
One example cited in the study, it includes a simple mathematical asking how much the kiwi is harvested in a single person over the course of several days. When he was presented with the details of the minor with respect to the size of some of the kivit, the corrected results the total of the final, even though the additional information does not have an impact on the solution.
This delicately on the justification, then, led the researchers to conclude that “the mere change of a name, can you change the outcome,” a sign of potentially serious concern for the future of the application and the one that require explanation for a stable and precise in the context of the real world.
According to the study, all of the models tested, the versions most of a small open-source as well as Called to the owner, as well as GPT-4o of OpenAI, told a sensible deterioration of the performance before the change in the appearance of a small data entry. Apple suggested that IT may be required to match the networks, the nerve on the grounds of the traditional one based on the symbols in order to achieve the ability to more accurate decision-making and the solution of the problem.
Discussion about this post