Stephen Miller gives clueless CNN employee a lesson in semantics.
CNN: “This is an undocumented immigrant...”
Miller: “They were here ILLEGALLY?”
CNN: “I want this to be a good-faith discussion...”
Miller: “When you use language designed to obscure the truth, that’s NOT good faith. An ILLEGAL ALIEN is an ILLEGAL ALIEN. They’re not ‘undocumented immigrants’.”
CNN: “Stephen, we’re trying to humanize the debate...”
Miller: “Humanize? Then let’s humanize the Penal Code too. The bank robber becomes a ‘specialist in unscheduled withdrawals’. The con artist turns into an ‘entrepreneur of alternative financial narratives’. The guy who breaks into your house uninvited is just a ‘visitor with pending documentation’. How about that?”
CNN: “That’s a crude analogy.”
Miller: “Crude is pretending federal law vanishes because CNN’s copy desk cooked up a cuddlier euphemism. ‘Undocumented’ sounds like someone who forgot their wallet in an Uber. No. It’s someone who crossed the border without authorization, stayed, and now the press treats the legal statute like it’s a matter of writing style.”
CNN: “But many contribute to the economy...”
Miller: “Ah, the classic argument: crime stops being crime if the person washes dishes or builds roofs. By the same token, the cigarette smuggler ‘contributes’ too, creates jobs in transport. Want us to call him an ‘informal logistics entrepreneur’? Would CNN go for that?”
CNN: “You’re polarizing.”
Miller: “No. I’m rejecting the marketing dictionary. You change the word, then accuse anyone who uses the legal term of being ‘inhumane’. It’s the oldest trick in the book: first soften the vocabulary, then declare moral victory. The United States is tired of this theater. Call it what it is. Or admit the goal was never clarity—it was narrative control.”
Tomorrow is a big day for the proposed $10M Spirit Airlines data deal. The bankruptcy hearing is at 11am ET.
Last week the flight attendant and pilot unions dropped their objections after changes were made to exclude sensitive data such as employee medical files, drug tests, and passport data. Other data such as emails, teams chats, and company documents are still in scope but must be de-identified.
Google is required to have a 3rd party perform the de-identification and test it according to ISO/IEC 27559. And then have a separate firm review samples to validate the de-identification process.
De-identification is table stakes nowadays. The hard part is keeping all the relationships across the data corpus with consistent replacements for each sensitive entity. If you can't do this, the data isn't going to be sellable.
Scammers are getting good.
Targeting VCs by sending emails about their Four Seasons accounts being changed.
Little do they realize that emerging managers don't have Four Seasons accounts 😎
Our directory of RL/data companies that buy and sell AI training data just had it's biggest update
208 companies now, up from 187 last week https://t.co/a3U4qcTvKz
This week: Mercor raised $750M at a $20B valuation
Who's buying is at 44 programs: @redpineai joins with a revenue share for rights holders
New on the list:
- @CuriosityStream (CEO @ClintLS): Nasdaq-listed, licenses video, audio and code datasets
- @redpineai (@andyhammarback@officemobile): licensed non-public research and data, €6.8M seed
- @psdnai (@storysylee): egocentric video and robot data from paid contributors, $15M seed led by a16z crypto
- @raycasterai: evals and environments for document and spreadsheet work
- Coarena by @coastyai (@prateekjannu@llmhubdev): computer-use preference data from a public arena
- @milliondotjs (@aidenybai@nisargptel): frontend coding RL envs and evals
- @panoptivehealth (@ymkallel): RL envs for clinical-trial operations
- @twinbaydotai: stateful copies of APIs like Stripe and HubSpot for training agents
- @autelica: real robot sites rebuilt as simulations
- @FSStudioDev: robot teleop and handheld-gripper data collection
- @medcasehealth: clinician-led medical annotation and evals
- @aitraining_jobs (@yumahey): expert evals, rubrics and RL tasks
- Archangel Health (@tej_xp): medical RL environments in a sandboxed EHR
- 3DCH Media (@TorstenHQ): licensed documentary video archives
- Also added: Algivra, CollectData, DecisionFacts, DeepLLMData, FasterLabeling, Logitly, Tiki AI
Who's missing?
As an experiment I asked Fable 5.2 to drive $10K in new MRR for Voiceflow - all on its own. No guidance, no human intervention.
To my surprise, it worked. Fully autonomous.
In less than 24 hours it was able to drive $10K in new MRR for Anthropic.
No jobs, no insurance for employees, no 401Ks for workers, no power, no innovation, no future.
You saved an abandoned, empty lot. And somehow you are proud of it.
This is an example of a policy that sounds good on paper (stick it to the credit card companies!) but will immediately create a ripple effect that politicians refuse to acknowledge. The cap will make it impossible to lend money profitably to a HUGE percentage of America. If this goes through, the people most in need of access to credit will just be cut off. They won't benefit from paying less. They'll just get cut out of the system.
This is the part most dashboards hide. Many failed payments are not lost customers. The card was fine, but the bank said no. That is a false decline, and you can often recover it with smarter retries. Our portfolio company Revaly (https://t.co/jXKHqy4cLq) does exactly that. Count recovered revenue separately from true churn, and your MRR gets honest fast.
Sometimes you see a graph that stops you in your tracks
October makes September look like a rounding error 👀
Never seen anything like this in my decade as an investor…
It’s remarkable that @OSFICanada and @FinanceCanada force Canadian banks to spend billions on AML practices, while turning a blind eye to the money laundering going on just across the street from said bank.