Language models are better than search engines, according to the Meta team. The researchers think this will be the next interface for accessing scientific knowledge.
Language models can possibly store, combine, and reason about information. It's important that that potentially is. It's an admission that language models can't do all these things. They might not be able to.
Shah says that language models aren't really knowledgeable beyond their ability to capture patterns of strings of words. A false sense of intelligence is given by it.
Gary Marcus, a cognitive scientist at New York University and a vocal critic of deep learning, gave his view in a Substack post.
The idea of language models replacing search engines is supported by many companies. For the last couple of years, Google has promoted its language model as a way to find information.
It is a very intriguing idea. It is irresponsible to suggest that the human-like text generated by such models will always be trustworthy. The error was not forced.
All Rights Reserved © 2024
