273 karma · joined July 29, 2021
But Microsoft is selling a service capable of such reproduction. They're selling access to an archive containing copyrighted code.
To me it's the equivalent of selling someone a DVD set of pirated movies. The DVD set doesn't "reproduce" the copyrighted material unless you "prompt" it to (by looking through the set to find the movie, and then putting it in your DVD player), but it was already there to begin with.
It is intuitively obvious that these problems would get even worse if the garbage output found its way into the training set, and not just into the context window.
Don't know if it would work but it's interesting.
This gets very hard to reason about because Arizona doesn't _do_ daylight savings. Like, what happens when we request 24 hours of data from them, on the day daylight savings time flips over? Do we want midnight-to-midnight data, or do we actually want 24 hours of data, which might wind up timestamped differently since there's a duplicate 2AM one day and a missing 2AM another day.
A few years ago we had a contractor write some extremely messy code to handle cases like that, and soon it's going to be my job to try and refactor it into something readable.
"I accidentally a coca cola bottle"
"accidentally what?"
"a coca cola bottle"
I think trying to model the world based on a single projection won't get you very far.
The same goes for curly quotes.
"Stacker" arcade machines work like that. It's always possible to win the minor prize (a little plastic toy generally), but if you refuse the small prize and keep playing for a major prize (game consoles, gift cards, etc.), it's impossible to win until the machine has banked enough money for the operator to make a profit.
2. Copyright shouldn't exist at all, but until it's abolished I am not opposed to a system where revenue is shared with the developers of any libraries you use. Although I think such a system would be difficult to manage and track.