3 ms·

In that world, high quality pre-AI texts might be come really valuable, much like low-background steel.

hinkley 3y ago
We might always have a certain volume of music and literature that can be training data, because even if it’s synthetic it’s still popular, which means it speaks to a subset of humans. That everyone reads the next Harry Potter indicates the impact of those 80k words. But we also know that kid who learned everything from books, pronounces the words wrong and uses definitions for them that nobody has used in decades (work with one of those now. I thought I could talk people to death, and he wears even me out.) those AIs will sound like out of touch nerds too.
- imdsm 3y ago
  Interesting perspective, but not all book learners mispronounce words or use outdated terms. Broad reading can expose us to different viewpoints and language styles. And re: 'out of touch nerds' – remember, they/we often bring groundbreaking ideas. Let's not undersell varied learning or language evolution.
  - hinkley 3y ago
    It’s a balance. An ex read every book in a small town library and spent a lot of college and early twenties relearning to say things right. By 28 she hardly ever got something wrong but if the subject ever came up she had plenty to say. Someone once ranted about people using big words, “having a crush on their high school English teacher they never got over.” I knew exactly what he meant. I spent my childhood hiding how smart I was and part of my 20’s reveling in it. It’s off putting. Wisdom comes from everywhere, and the smartest often have the least. Now I’m solidly in the Feynman camp: if you can’t explain your domain to college freshmen then you don’t know what you’re talking about (yet). When I’m refactoring code and introduce a new concept that’s like another one but different rules, I jump straight to a thesaurus. The word I pick out of the air might be good enough, but I guarantee you there’s a better one out there. Not fanciest or longest word, the most concise one. (Example: kind vs type in some circles of Type Theory). Some people act like that’s a crutch, but I’ve reviewed or refactored their code so I know that opinion and $5 isn’t worth a cup of coffee.
johnmw 3y ago
I like the sci-fi extension of this idea - pre-AI texts become as valuable as such steel, until they are able to successfully synthesize pre-AI data - in this case by running real world simulations. Do you really think it is just a coincidence that you happen to exist at the very last point of high value pre-AI data? ;-)
- hinkley 3y ago
  What is the Matrix?
BeFlatXIII 3y ago
Unlike low-background steel, the bits and bytes comprising the pre-AI corpus are infinitely copied. Their integrity is considered sacred, though that doesn't stop AI companies from attempting to pollute the training data of their competitors.

We might always have a certain volume of music and literature that can be training data, because even if it’s synthetic it’s still popular, which means it speaks to a subset of humans. That everyone reads the next Harry Potter indicates the impact of those 80k words. But we also know that kid who learned everything from books, pronounces the words wrong and uses definitions for them that nobody has used in decades (work with one of those now. I thought I could talk people to death, and he wears even me out.) those AIs will sound like out of touch nerds too.

imdsm 3y ago
Interesting perspective, but not all book learners mispronounce words or use outdated terms. Broad reading can expose us to different viewpoints and language styles. And re: 'out of touch nerds' – remember, they/we often bring groundbreaking ideas. Let's not undersell varied learning or language evolution.
- hinkley 3y ago
  It’s a balance. An ex read every book in a small town library and spent a lot of college and early twenties relearning to say things right. By 28 she hardly ever got something wrong but if the subject ever came up she had plenty to say. Someone once ranted about people using big words, “having a crush on their high school English teacher they never got over.” I knew exactly what he meant. I spent my childhood hiding how smart I was and part of my 20’s reveling in it. It’s off putting. Wisdom comes from everywhere, and the smartest often have the least. Now I’m solidly in the Feynman camp: if you can’t explain your domain to college freshmen then you don’t know what you’re talking about (yet). When I’m refactoring code and introduce a new concept that’s like another one but different rules, I jump straight to a thesaurus. The word I pick out of the air might be good enough, but I guarantee you there’s a better one out there. Not fanciest or longest word, the most concise one. (Example: kind vs type in some circles of Type Theory). Some people act like that’s a crutch, but I’ve reviewed or refactored their code so I know that opinion and $5 isn’t worth a cup of coffee.

johnmw 3y ago

I like the sci-fi extension of this idea - pre-AI texts become as valuable as such steel, until they are able to successfully synthesize pre-AI data - in this case by running real world simulations. Do you really think it is just a coincidence that you happen to exist at the very last point of high value pre-AI data? ;-)

hinkley 3y ago
What is the Matrix?

BeFlatXIII 3y ago

Unlike low-background steel, the bits and bytes comprising the pre-AI corpus are infinitely copied. Their integrity is considered sacred, though that doesn't stop AI companies from attempting to pollute the training data of their competitors.