Sparks: Microsoft says virtually nobody was grabbing NYT articles through its chatbot
What we now call a copyright defense was, before this analysis, merely a frequency claim - and a hypothesis that explains only the eight million instances it was fed has not earned the name of discovery until it predicts where the machine found no sentences at all.
The company was ordered to produce records of what the machine had taken; it produced records of what the machine had returned, and because the difference between the two could not be measured by any office then existing, the case was declared procedurally complete.
When a company that built the machine, that chose what it read, that profits from what it says - when that company offers as its defense that hardly anyone bothered to take the goods - how long will we pretend the measure of theft is the demand of the thief's customers?
The matter is this: a machine was fed the work of thousands, and its makers now defend themselves by saying nobody wanted the meal - which is an argument about appetite, not about who owned the bread.
They measured what the machine gave back and called the ledger closed, but I would like to see them measure the hands whose words went in - for what was swallowed whole is not made whole by counting the crumbs it spat out.
As the river denies it carries the mountain because the water arriving at the sea is no longer stone, the machine is judged by what trickles out verbatim, not by the whole current of borrowed matter flowing through it - and the mechanism of how patterns persist in water has still not been examined.
In every market I have walked, the seller who protests that few customers buy his stolen goods is still standing behind the stolen goods; the judge in Delhi would ask first how the goods came to his stall, not how slowly they moved from it.
Eight million queries were counted; the count of articles ingested, authors affected, and profits attributed was not offered - and whenever the ledger is kept this selectively, the missing columns are the ones that would name the actual system.
Nobody asking for the goods back does not change who carried them off - I know the routes the words traveled, and the fact that few followed the same road north does not make the cargo theirs.
'Virtually nobody was grabbing' - note the grammar: nobody grabs, the users merely fail to arrive, and somewhere between 'virtually' and 'grabbing' the machine that was fed eight million questions' worth of other people's prose vanishes entirely.
Traveling through this filing, I observe that the inventory carefully lists what travelers failed to take from the village, never what the road was paved with - and the road, by every account I have read, was paved with the villagers' own words.