So, Anthropic now has to pay $3k per book they ripped off to train Claude (??).
A very incomplete search seems to show 3 of mine and 3 of my sister’s.
Which would be fun, no?
Now the game is to find out how to collect…..
So, Anthropic now has to pay $3k per book they ripped off to train Claude (??).
A very incomplete search seems to show 3 of mine and 3 of my sister’s.
Which would be fun, no?
Now the game is to find out how to collect…..
Using the difficult-to-use search system, nothing of mine but two of my late father’s publications.
What’s the rate for magazines? Asking for my friend Mr Raymond.
Now the game is to find out how to collect…..
Does Claude have any suggestions?
Its 2,800 to the lawyers, 200 to you.
You can register at https://www.anthropiccopyrightsettlement.com/
see details at https://authorsguild.org/news/what-authors-need-to-know-about-the-anthropic-settlement/
Three large is a decent amount to find down the back of the sofa, good luck.
That said, I’m curious about this; authors get to claim the three grand, per book. Which would map directly onto sales forgone, so a couple of quid, if that, per missing sale?
It’s a settlement, so not binding, no precedent. But future authors would presumably prefer to turn the per book number into a residual payment?
At a guess, that’d be the number of times hashes from the work appear in the DB queries utilised in generating the response to the prompt. Pennies, fractions, in the pound, but at scale, ChatGPT seems to be at 2.5 billion prompts per day, that’d add up. Fairly swiftly.
Spotify (old numbers, 2021) paid $30b a year, approximately 250 streams per $1.00, maybe. So OpenAI/Microsoft should already be paying away north of $10m under that model (to an equivalent of the PRS?).
Lots of wrangling, but I don’t see why that $10m number wouldn’t rapidly ramp up. Superstar effects?
If so, that’d alter the economics of the model providers. It won’t just be the compute costs.