Tom WalkerLong story short, don't trust AI with your dog.
potpie@tomw Pretending that an LLM is an actual AI because it spits out pertinent and grammatical sentences, is like pretending a book is sentient because it contains recorded thoughts.
Jón Grétar@JonGretar @tomw That strikes me as a good point, but I like to drill down: there are 2 readings of "artificial," one which includes the meaning "fake" and another which only means "created" as opposed to "natural." An artificial dog would be a robot and NOT a dog. But an artificial sweetener is indeed a sweetener, just one that does not occur naturally. As long as we recognize that a fully conscious AI would be truly I and only A in the latter sense.
@potpie @tomw we can go deeper into that rabbit hole as the word “created” has it’s own problems. Do you “create” children? You certainly in a way program them by giving behaviour to learn from and try to guide and correct behaviour. Is the difference between humans and LLMs just the dataset, the fact that learning is turned off, and IO capabilities?
I think we give “conciousness” a little bit too much credit.
I think we give “conciousness” a little bit too much credit.
@JonGretar @tomw I do think we give consciousness too much credit, but also I look at the LLMs we have now and I see little beyond long range predictive text. Supporting point: their translations usually make no distinction between using different words and using different punctuation/capitalization; the input and the output are only connected by machine-learned spaghetti logic; there's no other layers of reckoning or interpretation. It certainly seems beain-like, but only like part of a brain.