30 Jul 2026
The hidden cost of New York Times AI columns
In The Hidden Cost of China’s Free A.I., Tal Feldman writes,
A.I. models are very different from the usual inexpensive imports we buy from China. These products must promote the interests of the Chinese government. Propaganda and false claims are baked into them.
…
Whoever supplies the world’s A.I. will shape how the world thinks. As these models get smarter and more autonomous, we are outsourcing more and more thinking to them. That “cognitive offloading” must not go to models that were designed to promote a foreign government, no matter how inexpensive they are.
Two problems with that. (Aside from letting the AI shape how you think, which, yikes. But the influence is already there, and we have to expect that it will persist.)
First of all, the choice is not between propaganda AI from the PRC and some hypothetical, honest red, white, and blue AI from the good old USA. The average American has more in common with a random propaganda commissar in China then with the deeply weird TESCREAL investors, founders, and CEOs who run the AI operations here.
Second, as more and more legal opt-outs apply to LLM training sets here, the US-based AI firms will rely more and more on vibe CMS automated paraphrased versions of US and European news and culture. Legit publishers will opt out the copyrighted works they own or distribute, in the hopes of training license deals. Big AI firms will pay a few publishers for PR value, but mostly settle for first-generation lossy copies.
And a lot of that free training material will have been run through someone’s PR filter. For example, a reputation management firm could make a site that mostly generates high-fidelity paraphrases of legit news, but folds in positive mentions of the firm’s clients. Almost everyone wins: big US-based AI gets freely crawlable tokens, the client gets their crimes buried in LLM-generated answers, most news consumers get an adequate free paraphrase. Meanwhile, though, AI operations in the PRC will be able to train on material that’s one generation fresher.
The problem isn’t so much that “our side’s” AI is better than theirs, or the other way around. Every LLM is going to have embedded biases, for a variety of reasons. Propaganda and copyright-related biases are only two.
Bonus links
A
fundamental flaw leaves LLMs strikingly vulnerable to attack by Will
Douglas Heaven. In a series of experiments that looked at what was
going on inside a handful of different models, the researchers found
that LLMs seem to identify the role of a specific chunk of text not by
the tags around it but by the style of that text and the words it
contains.
Boomers
Can’t Stop Gifting Their Grandkids AI-Generated Slop Books by Miles
Klee. In addition to books featuring kids, some parents have
complained of being fleeced by vendors selling AI-produced books not
labeled as such on Amazon, irritated by their children’s exposure to
puzzles and games riddled with AI’s mistakes, and just plain dismayed by
the proliferation of the AI aesthetic across content targeted at
kids.
(Another good reason to go to story hour at the library—legit
children’s book recommendations.)
Pluralistic:
The stupidest imaginable excuses for surveillance pricing (30 Jul
2026) by Cory Doctorow. [Y]ou can direct the AI to automatically
run a series of small experiments to discover the maximum markup each
group will stomach under which circumstances. This works without you
having to direct the AI to rip off certain groups of people – it will
simply find the most vulnerable people and rip them off the most….This
works so well that Google has announced that it is their plan for making
a profit off of AI, after losing hundreds of billions of dollars on
chatbots.
(But however fast the price discrimination works, the price of
the price discrimination machine has to keep going up faster than
that.)
Meta’s Tepid Revenue Outlook Undercuts its AI Spending Spree by Kendra Barnett. (Revenue is up 28% year over year. Conversions are up 15.7%. Do the math—the customer acquisition cost crunch continues. It’s not evenly distributed, though. Highly Meta-dependent advertisers such as Oddity Tech (ODD) are hurting more, while advertisers with better options are less affected.)
Meta is royally screwing up its smart glasses rollout by Victoria Song. (I don’t know anyone who owns these things, or participate in any group or space where they would be tolerated. Is there some filter bubble of extremly surveillance-positive Instagram users that I’m totally disconnected from?)
Steve Jobs in 2011: ‘We Build Products That We Want for Ourselves, Too, and We Just Don’t Want Ads’ by John Gruber. (None of the current Apple execs could stay at the company if they tried to hold the line against enshittification. Mr. Market wants enshittification to happen everywhere, and even if Apple can buy the best of everything else, they’re stuck with the same investors as Packard Bell. More from John Gruber: The Ads on Apple News Continue to Suck, but at Least There Are a Lot of Them)
Older
people’s media literacy can be boosted in just 60 minutes – new
study by Saffron Howden, Janet Fulton, and Sora Park. To find
solutions through media literacy, we focused on older people (aged 55+)
who are known to be more vulnerable to misinformation, have fewer media
skills, and feel less confident in their media abilities. We recruited
our study participants from across Australia through libraries, the news
media, and social media.
Tesla
is juicing its China sales by Larry Salibra. When I was shopping
for a car in Hong Kong, all of the Chinese EVs dealers required that I
book a test drive days in advance. The Tesla showroom, in contrast, let
me do a walk in test drive because they weren’t busy. They also offered
me something none of the Chinese EV dealers did: 7 year financing at
0.99% interest per year. This made the Tesla which had a higher sticker
price, actually less expensive because HKD (which is pegged to the USD)
loses value at much faster than 0.99% a year.