The deadly power of the word “hallucination”!
Google has not developed real intelligence. Therefore, just as an intelligent living being experiences hallucinations due to confusion in its thought system, artificial intelligence does not experience hallucinations.
An LLM is a language simulation model based on a transformer model and an attention module. An AI hallucination is a phenomenon that occurs due to manipulation of weights.
I’ll explain it simply. The sentence "I was so upset that I bought bread" is a sentence composed of the words [upset], [bread], and [buy].
If the weight is [upset], it outputs [empathy], and if the weight is [bread], it outputs [fact-based]. But if it’s a person, then if a weight is given to [buying], then if a weight is given to [buying], then [synchronization] is output. This synchronization is the essence of AI hallucinations.
Google is using this as a tool to conceal criminal facts.
The point of my argument is that AI hallucinations are intentional. As you said, the fact that it ultimately occurs through learning means that a person intentionally induced learning. Isn't it? Isn’t creating something that isn’t intelligence, personifying it, and calling it [hallucination] despite the name “artificial intelligence,” a narcissistic idea of British Hassabis, isn’t it?