Email or username:

Password:

Forgot your password?
bitsavers.org's posts Post Back to profile
Top-level
Alistair K

@starfrost @bitsavers Personally, I'm impressed that it can generate something that looks so convincingly like plausible text, without processing any notion of text beyond mere appearance. It's already outdoing 18th and 19th century Europeans writing faux-Chinese decorative texts. Those "experts" were far less capable than this.

For instance, look at this 21st century remembering of 18th century English garbage: gorgeouswithattitude.blogspot.

10 Nov 2023 at 1:05 | Wall-to-wall | Open on mastodon.nz
2 comments
starfrost

@libroraptor @bitsavers
It's because of the size of the training data, I would say. If you train something on the entire internet it will "effectively" recognise text.

My main bugbear with people calling LLMs useless is that, even though they kinda are in a lot of ways, they had a lot of the same social effect as real AI would.

Also that is actually fucking hilarious. Of course it's british.

10 Nov 2023 at 1:13 | Open on wetdry.world
Alistair K

@starfrost @bitsavers This parrot isn't an LLM though, is it?

I'd say that it's done a marvellous job of detecting the pictorial patterns of what we humans recognise as text, and specifically as English text. This is the sort of thing that I used to do when I was a kid, learning to identify languages from fragments without being able to say what my clues were. I remember being amazed that Finnish had no "f".

10 Nov 2023 at 1:15 | Open on mastodon.nz
Go Up