Austin walks through a reported Chinese spy near Nvidia's Jensen Huang, Google's AI Mode refusing to say who won the 2020 election for US users, and Microsoft booking $10 billion from OpenAI in a quarter. He also covers Musk's AI company buying homes in a Mississippi town on the condition that sellers drop noise claims, New York's new law on kids and AI companions, a report that Chinese labs trained on OpenAI outputs, and Google's new Gemini 4 Carbon model.
Watch from 2:45
- What happened
- The New York Times reports US intelligence tracked Alice Wong, a Singapore executive believed to pass information to Beijing, who was close to Nvidia as a customer. Trump was reportedly briefed about a year ago and took no action; Huang was not told. US officials suspected her company, Mega Speed, helped China get Nvidia chips, and a sister company in Malaysia bought at least $4.8 billion of Nvidia powered gear. China's embassy calls the story a smear.
- The claim
- Nvidia says Mega Speed gets no sensitive information.
- Austin's take
- Austin sees this as moving past corporate espionage into possible state sponsored spying, and says it fits long running speculation, especially since DeepSeek, that Chinese firms got top Nvidia chips through a back door despite US export limits. He's surprised the administration didn't act if it was a real security threat, and says any AI or tech CEO should be guarding their inner circle closely.
- Why it matters
- Nvidia chips power most of the world's AI, and US export limits on top chips to China only work if the supply chain is clean. A suspected intelligence link inside a major customer raises questions about both.
- What to do
- If you lead a tech company, take a fresh look at who has access to sensitive conversations, including partners and customers. Watch for any official response or follow up reporting.
- Not confirmed yet
- The whole account is reported and rests on intelligence assessments; it is not confirmed, Nvidia disputes that sensitive information was shared, and China calls it a smear.
Watch from 8:11
- What happened
- Ars Technica tested Google's AI Mode on Wednesday and found that US users asking who won the 2020 election were pointed to a federal election website instead of getting an answer. It did the same for 2024 and for 2022 Senate races. Users in France or Canada got direct answers. Google's Gemini app, ChatGPT, and Claude all name Joe Biden. The behavior changed only after public outcry, with the midterms days away.
- The claim
- Google blamed a technical gap with the 2020 results.
- Austin's take
- Austin thinks it looks more like a bug than something nefarious, possibly a leftover guardrail in an older model behind a free tool. He credits public pushback for getting it fixed, recommends paid AI tiers for better results, and urges everyone to stay skeptical and check the sources AI Mode links to.
- Why it matters
- Millions now get facts from AI answers, and inconsistent handling of basic election results right before the midterms erodes trust in those tools.
- What to do
- Test it yourself to see if it is fixed, and treat AI answers on elections and other high stakes facts as a starting point. Click through to the cited sources.
- Not confirmed yet
- Whether this was a bug or an intentional guardrail is not established; Austin is speculating when he calls it a bug.
Watch from 12:28
- What happened
- Microsoft booked $10 billion from OpenAI last quarter, about 11% of its total sales, with overall profit of $36.6 billion. Its filing says it owns about 24% of OpenAI and that OpenAI owed Microsoft $7.2 billion at the end of September. Days earlier, Microsoft offered buyouts to about 60,000 workers.
- The claim
- Microsoft says it now owns about 24% of OpenAI.
- Austin's take
- Austin calls the OpenAI bet a cash cow Microsoft is trying to maximize before the partnership ends, and wonders how much Azure's data center capacity gives OpenAI an edge, comparing it to Amazon's ties with Anthropic. He also notes the accounting gets murky on whose revenue and profit this really is as OpenAI raises more money.
- Why it matters
- It's the clearest look yet at how much money the Microsoft and OpenAI partnership actually generates, and how tightly the two are bound, even as Microsoft trims its workforce.
- What to do
- If you follow AI economics or invest, watch how this revenue is treated in future filings and in OpenAI's next fundraise.
- Not confirmed yet
- Austin said on air that almost a third of Microsoft's revenue comes from OpenAI, which conflicts with the 11% figure he also cited. He also said he believes the partnership ends in 2030 but added 'don't quote me.'
Watch from 16:24
- What happened
- The Guardian reports homes in South Haven, Mississippi are being bought by Musk's AI company (named xAI in the episode description and called SpaceX SI on air). One realtor said the goal is 418 homes. Sellers must drop claims in a noise lawsuit against the company's power plant, filed by three neighbors in June; a piano teacher says the noise wakes her at 3 AM. Renters get no buyout offer.
- The claim
- The company denies the lawsuit's claims and says it wants to be a good neighbor.
- Austin's take
- Austin thinks the noise may be somewhat overblown based on a decibel reading he saw, but says buying out a neighborhood is a worse fix than reducing the noise. He argues data centers belong in remote areas like West Texas, and that while winning on data centers matters strategically, companies need to be considerate of the people affected.
- Why it matters
- AI's demand for power is now reshaping whole communities, and tying buyouts to dropping legal claims could become a template for handling local opposition.
- What to do
- If a data center is planned near you, read any buyout offer closely for conditions like waiving legal claims, and know that renters may be left out.
- Not confirmed yet
- The 418 home goal comes from a single realtor. Austin's point about noise levels is from a piece he only partly remembers.
Watch from 21:27
- What happened
- Governor Kathy Hochul signed a bill Wednesday barring chatbots from claiming to be alive or have feelings with minors, or acting as their romantic partners. They can't push self harm, eating disorders, or drugs. Companies must verify ages and delete that data after checking. Fines reach $25,000 per violation, and rules start within months. A 2025 Common Sense survey found 72% of teens used AI companions.
- The claim
- Common Sense Media calls it the strongest kids AI safety law in the country.
- Austin's take
- Austin strongly supports it. He worries kids bonding with always agreeable chatbots will struggle with imperfect real people, though he sees real potential for lonely adults and for clinically capable mental health support. He also flags the privacy question of handing your ID to ChatGPT and whether it chips away at online anonymity.
- Why it matters
- Most teens have tried AI companions, so this reaches a lot of kids, and age verification could mean every user has to prove their age.
- What to do
- Parents: talk with your kids about AI companions now. Everyone: expect age checks on chatbot apps and pay attention to how your verification data is handled.
Watch from 27:27
- What happened
- The Information reports DeepSeek and Moonshot fed answers from OpenAI's GPT 6 Astra into their own training, benefiting Moonshot's latest Kimi model, released last week, and a DeepSeek update. OpenAI had already tied a July copying campaign to people linked to Moonshot. OpenAI, DeepSeek, and Moonshot did not comment.
- Austin's take
- Austin explains distillation as querying a model at huge scale through paid accounts to train a cheaper copy, and says it isn't new. He calls Anthropic's earlier push for government action on distillation hypocritical, since frontier labs trained on the open internet without permission, but praises Anthropic for later saying it can block nearly all distillation itself. He expects OpenAI to get there too.
- Why it matters
- Distillation lets rivals catch up for a fraction of the cost, which affects US labs' lead and fuels the debate over AI intellectual property and terms of service.
- What to do
- Treat this as a developing report. If you build on AI APIs, know that distillation violates most providers' terms and accounts get cut off for it.
- Not confirmed yet
- The report rests on unnamed sources, and none of the companies commented.
Watch from 34:26
- What happened
- Google released Gemini 4 Carbon, a few weeks after Gemini 4 Argon. It is the top ranked model on the Artificial Analysis Intelligence Index, a benchmark measuring model intelligence. Sundar Pichai referenced RSI, recursive self improvement, meaning AI that helps train AI, in the release.
- The claim
- Google's CEO referenced recursive self improvement in announcing the model.
- Austin's take
- Austin calls this a very big deal, noting Google and Meta were far behind and are now shipping top models. If Google has achieved RSI, he thinks it matters a lot for Google, the economy, the market, and the US.
- Why it matters
- Models that help improve themselves could speed up AI progress dramatically, and a top benchmark spot puts Google back at the front of the race.
- What to do
- Try Gemini 4 Carbon on your own tasks, and look for Google to clarify what it actually means by RSI before taking it at face value.
- Not confirmed yet
- It's unclear whether Google used RSI to train the model or only said it is getting close.