James Yu

Wanting to strictly police how other people use their compute is fascist behavior

Have you ever thought you understood something and then it did something that you thought was impossible?

One can be surprised by things you understand. I actually find that joyful.

is the set of people who you find credible have a zero intersect with those that are impressed with llms? the truth is likely in the middle, given how widely used (and rententive) these models are. there are things they are not good at, and there are things they are impressive at.

this exactly. every few months, there has been a leap where LLMs can do something previously thought impossible. just ask any coder that has seriously used LLMs - they have thoroughly exceeded many expectations on a daily basis

False. We have never trained on authors data without consent.

he did do that: www.bfi.org.uk/sight-and-so...

I've corrected you on this many times already. We've never "harvested" fanfic. We use all the big models out there (GPT, etc) and when we do fine-tune our own models, it's using data with consent.

Purely AI-generated art can’t get copyright protection, says Copyright Office AI prompts are different from Jackson Pollock’s paintings, apparently.

This is the right ruling! If an artist or writer uses AI as part of a larger process copyright protection should be granted www.theverge.com/news/602096/...

This would be taken up case by case, as it always has been

I believe this will happen soon. But it depends on what you mean by interesting?

We have some stuff in the works where for the first time, I’m seeing truly interesting (imo) short stories emerging. Which is quite a leap from 2 yrs ago. DM if you’re interested in seeing

Japan is liberal about tearing down old buildings to make way for the new.

Japan is liberal on AI and copyright, allowing training without restrictions.

These feel connected.

Was there election interference in 2020?

My pet peeve is when ai creativity studies use gpt 3.5 turbo that involve human feedback. That’s two generations behind. It's like doing a study on video game graphics and asking people to play pong

is anyone building an abstraction for rich text editing in browsers as outlined here (that is not dependent on content-editable)?
https://writer.zohopublic.com/writer/published/djw8l10ba521e51a34c2488530291ebf76dbf

Hey Paolo I’d love to chat with you about this. We aren’t striving to create the helicopter (and our writers aren’t using it like that). We think of it as a diverse set of vehicles for lateral thinking up the mountain, with as many twists and turns as non-AI assisted narrative writing

The whole wordcel vs shape rotator thing is a psyop to divide us: we are all symbol shifters

Word Rotators ftw

The Metamorphosis but Gregor wakes up as an LLM

Interesting WGA is taking issue with AI generated source material. Is this primarily a credits thing? Or something else?

“Do you know Python?”

“I AM Python!!!!”

First order of business: eradicate js developers

It’s hilarious to me that AGI will most likely be written in python

And that typography 🥰

Frog and Toad on apple+ is one of the most faithful adaptations of a children’s story that I’ve ever seen. And it’s the most wholesome thing since Bluey, something the world needs

Same with pandas

One striking thing that I learned at my time at Facebook was that if they had let the feed algo loose and not tuned it, most of the feed would have been videos of police car chases and polarizing local news, overwhelming friend content.

Says a lot about engagement

Hello 👋

I’m James, a friendly maker that loves to build tools to make your life better. I’m currently building Sudowrite, an AI assistant for narrative writers.

Excited to type many strings here. ✍️📝