Everyone has their own personal views on the appropriate terminologies to use for emerging tech. I remember having a huuge argument with everyone (at Tech2) about whether or not what Prisma does is apply filters, it was only later that the term 'style transfer' became commonplace. I was adament that they are not filters because the same photo processed with the same settings does not produce the same results twice.
I still hate the fact that image macros, or pictures with text on them, are called memes, and that the original definition, as coined by Dawkins in The Selfish Gene, has actually been lost. Cannot fight against everyone over it.
Even in Clubhouse when I talk to people about ChatGPT, I got interrupted when I said that you can reason with it to get around restrictions (was told its not a person, so cannot be reasoned with), and when I pointed out that it comes up with fake info without any basis, was told it hallucinates!
Anyway... what I was really saying is that it kind of mirrors you and your prompts, and throws stuff that you type back at you. For example, if you ask it how automobile manufacturing companies have been attempting to make wireless charging pads for electric vehicles for over a century now, it will throw up something like Toyota came up with some tech (experimental, first developed in 2010s) back in 1800. I also asked for other animal analogies used in quantum physics such as Schrodinger's Cat and Butterfly Effect.... and it came up with a ridiculous zoo of quantum flamingos, alligators and penguins, along with reasonable explanations, that just have not actually been used anywhere before. That is what I referred to as hallucinations, which is not lying, that is providing fake information with an intent to mislead, it is rather just producing novel terminologies from the foam of text corpora that it processes.
I really suspect that there are copyrighted books in the material that it has been trained on, because some of the information cannot be verified on anything available online, but I was able corroborate with books behind paywalls and meatspace. OpenAI has not publicly disclosed the text that ChatGPT has been trained on.
