most of the time you'll be talking to a bot there without even realizing. they're gonna feed you products and ads interwoven into conversations, and the AI can be controlled so its output reflects corporate interests. advertisers are gonna be able to buy access and run campaigns. based on their input, the AI can generate...
ChatGPT isn’t really as smart as a lot of us think it is. What it excels at really is just formatting data in a way that is similar to what you’d expect from a human knowledgeable in the subject. That is an amazing step forward in terms of language modeling, but when you get right down to it, it basically grabs the first google search result and wraps it up all fancy. It only seems good at deductive reasoning if the data it happens to fetch is good at deductive reasoning.
Chatgpt doesn’t actually understand language. It learns patterns in data it’s been fed (human generated language) and uses that to generate new, unique data which matches those patterns according to the prompt. In other words, it’s not really “thinking” in that language.
We understand spelling as a part of language - putting together letters to create words, then forming sentences according to a context. Chatgpt can’t do that since it doesn’t know how to speak English, only how to follow a list of instructions to form what appears to us as coherent English.
How is that possible? There’s such an easy model if one wanted to cheat the system.
ChatGPT isn’t really as smart as a lot of us think it is. What it excels at really is just formatting data in a way that is similar to what you’d expect from a human knowledgeable in the subject. That is an amazing step forward in terms of language modeling, but when you get right down to it, it basically grabs the first google search result and wraps it up all fancy. It only seems good at deductive reasoning if the data it happens to fetch is good at deductive reasoning.
…so, basically, it’s like an SEO-optimized “article”?
Chatgpt doesn’t actually understand language. It learns patterns in data it’s been fed (human generated language) and uses that to generate new, unique data which matches those patterns according to the prompt. In other words, it’s not really “thinking” in that language.
We understand spelling as a part of language - putting together letters to create words, then forming sentences according to a context. Chatgpt can’t do that since it doesn’t know how to speak English, only how to follow a list of instructions to form what appears to us as coherent English.
It also can’t play hangman for the same reason.
Check out the Chinese room argument.