Fair use is what enables quote-dunking. It's what lets you make fun of elected officials by quoting their own words back at them. It's what allows gifs to exist in the first place. It's what makes software interoperable. It underpins literally everything online.

We discovered something counterintuitive: the best way to find policy ambiguities is to create subtly distinct versions of your policy, then see where AI interprets them differently. Those mismatches reveal your problem spots. Systematic, repeatable, effective. But still manual. 🧵 4/7

even worse with the visual

trans catgirl hacker post is still up though 😎😎😎 x.com/grok/status/...

It seems like it pulls account info and recent interactions (that’s how it got Sutskever and EleutherAI if you look at previous responses to the same user)
But yeah it might also steer that way anyway if you go the “just asking questions” route

(I censored the photo, it’s Joe Biden saying the n-word)

the guy (bot?) that got the Hitler response looks like this, so yeah seems tailored to the user

ty for explaining the joke to the audience, I am still pondering the question, give me a few minutes

hm yes let me run “transgender catgirl hackers at EleutherAI” through our PR department

yeah you can still get it to endorse transgender catgirl hackers at EleutherAI

“groyper free association” is one way to describe this

please no more jokes allowed

Plus other observations: for example, in terms of text volume, open source code far outweighs other categories of "openly licensed" content. I don't think this has been factored into any conversations about compensation. Turns out it was about paying open source maintainers all along!

AI-generated images in menus on food delivery apps were the “please stop” moment for me

The project of open science for machine learning only works if we are able to distribute the training data. Openly licensed data lets us do that, under mild conditions. We make sure to provide document-level metadata for authorship, licensing information, links back to the originals, and more.

(1) reuse of which artifact associated with the project? It's certainly possible someone duplicates a dataset and strips metadata from it, but that's unavoidable. Part of the work was also promoting tooling, cc @storytracer.org blog.mozilla.org/en/mozilla/a...