(C) Alec Muffett's DropSafe blog.
Author Name: Alec Muffett
This story was originally published on allecmuffett.com. [1]
License: CC-BY-SA 3.0.[2]
Kevin A. Bryan on X: “NYT/OpenAI lawsuit completely misunderstands how LLMs work, and judges getting this wrong will do huge damage to AI. Basic point: LLMs DON’T “STORE” UNDERLYING TRAINING TEXT. It
2023-12-29 11:22:52+00:00
I’m seeing lawyers and their peers in Civil Society saying stuff along the lines of:
Ooh, the GPT4 bot is quoting entire NYT articles, and it’s leaving out stuff which reflects the NYT style guide which proves that it’s up to no good!
…which is an understandable position to take if and only if GPT4 possessed an actual consciousness. But the thing is: it’s not, and it doesn’t. The bot is not attempting to recite an NYT article from memory, instead it’s at the whim of the words it was prompted with setting it off down a path of repeating the statistically most likely words — a selection which is biased by virtue of the article being quoted and reposted widely around the internet — so that what comes out is an approximation of the article in its entirety.
Want to play the same game?
a rolling stone gathers […]
a bird in the […]
I now pronounce you […] you may now […]
See if you can complete all those sentences. You can probably come up with some plausible, if not certain, possible solutions — although those solutions will also reflect your personal biases, especially in the third example.
That’s what’s going on: GPT4 is just a massive engine for filling in the dots, spewing words into the void until it seems right, normal and typical to stop.
This thread from Kevin, if you have a Twitter account to read it, says it better.
[END]
[1] URL:
https://alecmuffett.com/article/108755
[2] URL:
https://creativecommons.org/licenses/by-sa/3.0/
DropSafe Blog via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/alecmuffett/
