Tech giants upset to find themselves on the short end of the same T&C/EULA stick they've subjected their own customers to for years.
An impressive ouroboros of hypocrisy:
- From treating users like rich ground to be fracked for data.
- To scraping the entirety of the public and private (wherever possible) internet to sell it back.
- To complaining that their competitors do the same to them via distillation.
- Now whining that the few AI companies producing anything genuinely useful (such that they feel compelled to pay for their tokens) are data fracking them right back.
It’s not even the act that bothers me, it’s the self-righteousness. It’s the relentless attempts at regulatory capture with real or pretend messiah complex.
You don’t have to design your product to allow distillation. Do your best to prevent it. And if they beat you, well, they beat you.
I suppose if the messiah complex is real being beaten means it’s the end of the world.
Life was good when only one group cried rapture every few years.
Nadella Blasts AI Industry's Double Standard: Allowed to Scrape Data, Yet Restricts Others from Distilling Models
Microsoft CEO Satya Nadella published a lengthy essay on X titled "The Reverse Information Paradox," sharply criticizing major AI model providers like OpenAI and Anthropic for a glaring double standard: they leverage public data and user interactions to train their own models while imposing restrictive terms that prevent other companies from distilling capabilities from those models.
He warned that if the learning process flows only one way—toward model providers—economic value will increasingly concentrate among the handful of firms controlling AI infrastructure. The debate intensified as Palantir CEO Alex Karp delivered an even more blistering critique on CNBC, claiming enterprise clients are "furious" over token-based pricing models.
Former White House AI czar David Sacks also joined the fray, accusing Anthropic of an "observe-copy-expand" pattern that erodes application-layer businesses. While critics note that both Nadella and Karp have their own commercial motives, the issues they raise—ballooning token costs and one-way knowledge flows—have become central anxieties in corporate AI strategy.
> Claude Mythos Preview was trained on a proprietary mix of publicly available information from the internet, public and private datasets, and synthetic data generated by other models.’
'synthetic data generated by other models.’
is it distillation ?
Tech giants upset to find themselves on the short end of the same T&C/EULA stick they've subjected their own customers to for years.
An impressive ouroboros of hypocrisy:
- From treating users like rich ground to be fracked for data.
- To scraping the entirety of the public and private (wherever possible) internet to sell it back.
- To complaining that their competitors do the same to them via distillation.
- Now whining that the few AI companies producing anything genuinely useful (such that they feel compelled to pay for their tokens) are data fracking them right back.
It’s not even the act that bothers me, it’s the self-righteousness. It’s the relentless attempts at regulatory capture with real or pretend messiah complex.
You don’t have to design your product to allow distillation. Do your best to prevent it. And if they beat you, well, they beat you.
I suppose if the messiah complex is real being beaten means it’s the end of the world.
Life was good when only one group cried rapture every few years.
Nadella Blasts AI Industry's Double Standard: Allowed to Scrape Data, Yet Restricts Others from Distilling Models
Microsoft CEO Satya Nadella published a lengthy essay on X titled "The Reverse Information Paradox," sharply criticizing major AI model providers like OpenAI and Anthropic for a glaring double standard: they leverage public data and user interactions to train their own models while imposing restrictive terms that prevent other companies from distilling capabilities from those models.
He warned that if the learning process flows only one way—toward model providers—economic value will increasingly concentrate among the handful of firms controlling AI infrastructure. The debate intensified as Palantir CEO Alex Karp delivered an even more blistering critique on CNBC, claiming enterprise clients are "furious" over token-based pricing models.
Former White House AI czar David Sacks also joined the fray, accusing Anthropic of an "observe-copy-expand" pattern that erodes application-layer businesses. While critics note that both Nadella and Karp have their own commercial motives, the issues they raise—ballooning token costs and one-way knowledge flows—have become central anxieties in corporate AI strategy.
> Claude Mythos Preview was trained on a proprietary mix of publicly available information from the internet, public and private datasets, and synthetic data generated by other models.’
'synthetic data generated by other models.’ is it distillation ?
> is it distillation
Yes. Though it's probably mostly just other Claudes.
Oh dear, imagine what the writing of a Claude trained on other Claudes is going to look like. I’m not sure I can handle that much Claudosity.
Claudception
Non-X Source blog post: https://snscratchpad.com/posts/reverse-information-paradox/
Some more discussion: https://news.ycombinator.com/item?id=48904480