I really enjoyed SN 911's title subject, A Clever Regurgitator. It was exactly what I was looking for, even though I didn't know it at the start. I'd been thinking of requesting a propeller hat, deep-dive into AI. But there are plenty of papers and lectures out there already (several of which are waiting for me in open Tabs), some books in my Humble Bundle library that I didn't know were there and a freshly purchased copy of Gilbert Strang's, Linear Algebra and Learning from Data (which I'll probably never read. It goes along with MIT OCW 18.065).
Anyway, Regurgitator really cut through the noise. The parrot was a perfect contrast to the large language AI. That ChatGPT simulates intelligence because it's absorbed the "historical global output of a truly intelligent species – namely, man ..." , it's able to say the right thing, "because that right thing has been expressed before, by man, in thousands of different contexts." (SN 911 Show Notes) Much has been made of ChatGPT's ability to write essays. But how many high school essays are really original? You earn a PhD for adding to the body of human knowledge. That's supposed to be the requirement, right? Not just stringing together a few quotes from Wikipedia and a couple of library books.
The same thing goes for translation, which probably should be better expressed as interpretation. Real-time 'translation' like they do with speeches at the UN for example, is technically interpretation, and seems like the perfect place for AI already. Translation, on the other hand often involves pondering over a complex passage from a text written for a foreign audience, trying to capture the author's idea and then convert that idea into text that the local audience will grasp (I'm looking at you, Thomas Mann
). It's been a while since I did any real translation, but in the past google.translate would do an excellent job with run-of-the-mill text, but whenever I was really stuck, it wouldn't give me anything that I could be sure was right. (I learned from the abstract of All You Need Is Attention (Attention ... Need--AI wouldn't have made that mistake) that German and French were the languages used to test many AIs. Presumably, large quantities of human translated works have been fed to those systems.) And syntactically, 'right', and for the author/translator 'right', may not always be the same thing. AI definitely speeds translation work, but I'm not going to bet my life on an artificially produced translation just yet.
So, Thanks Steve, for the reality check. A couple places I hope AI will someday take us: I'd like to see an AI interpret animal communication furthering Temple Grandin's work (I think my cat knows more than she's saying). And (this is being done--but can only get better) in the realm of brain-machine interfaces, more complex motor signals will be processed to produce finer movement (i.e. natural gait, fine hand motor) in assistive devices for the disabled.
Anyway, Regurgitator really cut through the noise. The parrot was a perfect contrast to the large language AI. That ChatGPT simulates intelligence because it's absorbed the "historical global output of a truly intelligent species – namely, man ..." , it's able to say the right thing, "because that right thing has been expressed before, by man, in thousands of different contexts." (SN 911 Show Notes) Much has been made of ChatGPT's ability to write essays. But how many high school essays are really original? You earn a PhD for adding to the body of human knowledge. That's supposed to be the requirement, right? Not just stringing together a few quotes from Wikipedia and a couple of library books.
The same thing goes for translation, which probably should be better expressed as interpretation. Real-time 'translation' like they do with speeches at the UN for example, is technically interpretation, and seems like the perfect place for AI already. Translation, on the other hand often involves pondering over a complex passage from a text written for a foreign audience, trying to capture the author's idea and then convert that idea into text that the local audience will grasp (I'm looking at you, Thomas Mann
). It's been a while since I did any real translation, but in the past google.translate would do an excellent job with run-of-the-mill text, but whenever I was really stuck, it wouldn't give me anything that I could be sure was right. (I learned from the abstract of So, Thanks Steve, for the reality check. A couple places I hope AI will someday take us: I'd like to see an AI interpret animal communication furthering Temple Grandin's work (I think my cat knows more than she's saying). And (this is being done--but can only get better) in the realm of brain-machine interfaces, more complex motor signals will be processed to produce finer movement (i.e. natural gait, fine hand motor) in assistive devices for the disabled.
Last edited:
