Thread by @CharlesRollet1

Scoop: @BusinessInsider obtained an internal list of websites that could and couldn't be used for training Anthropic's latest AI models.

Anthropic's contractor Surge AI left the list fully public on Google Docs.

'Sites you can use' include Bloomberg, Harvard, & the Mayo Clinic

Check out the list here:

Many of the whitelisted sources copyright or otherwise restrict their content.

At least 3 - the Mayo Clinic, Cornell University, & Morningstar - told BI they didn't have any AI training agreements with Anthropic.s3.documentcloud.org/documents/2602…

The spreadsheet also includes a blacklist of websites that Surge AI's gig workers were "now disallowed" from using.

The blacklist includes companies like the NYT & Reddit which have sued AI startups for scraping without permission.

Full doc here:
documentcloud.org/documents/2602…

The spreadsheet was used by Surge AI workers helping teach Anthropic's latest AI models be "helpful, honest, & harmless."

This was for RLHF purposes, not pre-training.

But legally, it's "probably not going to make a material difference in terms of fair use" a law prof told BI

Anthropic said it wasn't aware of the list and that Surge AI created it. (Surge declined to comment on this point.)

Surge locked down dozens of files for the project shortly after BI reached out & said it's "looking closely" into the security lapse.

For more, read our @BusinessInsider article: businessinsider.com/anthropic-surg…