Researchers at Apple Computer Company have found evidence, via testing, showing that the seemingly intelligent responses given by AI-based LLMs are little more than an illusion. In their paper posted on the arXiv preprint server, the researchers argue that after testing several LLMs, they found that they are not capable of performing genuine logical reasoning.
Over the past few years, many LLMs such as ChatGPT have developed to the point that many users have begun to wonder if they possess true intelligence. In this new effort, the team at Apple has addressed the question by assuming the answer lies in the ability of an intelligent being, or machine, to understand the nuances present in simple situations, which require logical reasoning.
One such nuance is the ability to separate pertinent information from information that is not pertinent. If a child asks a parent how many apples are in a bag, for example, while also noting that several are too small to eat, both the child and parent understand that the size of the apples has nothing to do with the number of them present. This is because they both possess logical reasoning abilities.
In this new study, the researchers tested several LLMs on their ability to truly understand what it is being asked, by asking them indirectly to ignore information that is not pertinent.
Their testing involved asking multiple LLMs hundreds of questions that have been used before as a means of testing the abilities of LLMs—but the researchers also included a bit of non-pertinent information. And that, they found, was enough to confuse the LLMs into giving wrong or even nonsensical answers to questions they had previously answered correctly.
This, the researchers suggest, shows that the LLMs do not really understand what they are being asked. They instead recognize the structure of a sentence and then spit out an answer based on what they have learned through machine-learning algorithms.
They also note that most of the LLMs they tested very often respond with answers that can seem correct, but upon further review are not, such as when asked how they “feel” about something and get responses that suggest the AI thinks it is capable of such behavior.
More information:
Iman Mirzadeh et al, GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models, arXiv (2024). DOI: 10.48550/arxiv.2410.05229
machinelearning.apple.com/research/gsm-symbolic
arXiv
© 2024 Science X Network
Citation:
Apple researchers suggest artificial intelligence is still mostly an illusion (2024, October 16)
retrieved 17 October 2024
from https://techxplore.com/news/2024-10-apple-artificial-intelligence-illusion.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
