Data centers lose 40 of 41 votes

what to know for now

📉 AI stocks had their worst day since April 2025, two days after the midterms. On Thursday, November 5, the Nasdaq Composite fell 5.5%, Nvidia fell 9.3% (wiping out its 2026 gain), and AMD and Micron each fell about 17% as investors priced in a Congress and a set of new governors who could slow the data center buildout. J.P. Morgan's Michael Cembalest also blamed "a heavy bout of hedge fund profit-taking." And AMD got 11% back the next day. Read more

🗳️ Data centers lost 40 of 41 local ballot measures, with voters picking pause-backing governors in Texas, Ohio and Georgia. Gina Hinojosa beat Greg Abbott after promising to halt every data center approval until the legislature passes new laws, Amy Acton won Ohio on a "conditional moratorium," and Keisha Lance Bottoms won Georgia promising a temporary pause by executive order. Democrats also took both chambers of Congress. 21 of those 41 local measures were in Ohio. Read more

🕵️ Nvidia’s board opened an independent investigation into Jensen Huang after reports that he was the target of Chinese intelligence. The Financial Times reported on November 7 that a classified US memo says a woman close to Huang was recruited by China’s Ministry of State Security, Huang told staff he had been unfaithful to his wife, and the board says it has “no reason to believe” national security or Nvidia secrets were compromised. The stock fell about 5% before Monday’s open, eight days before Nvidia reports earnings on November 17. Read more

🧠 MIT researchers had GPT-6 Astra edit a small model’s weights directly and lifted its math score from 7.9% to 48.1%. Changing 0.016% of Qwen3.5-0.8B’s weights mostly fixed a habit of looping until it ran out of room (93% of its answers hit the length limit before, 33.1% after), and the authors say the method “could not add an ability the model lacked.” It’s a blog post, not a peer-reviewed paper, and a smaller Gemma model barely moved. Read more

📈 Anthropic's updated IPO filing shows $17 billion in third-quarter revenue. It lost $567 million running the business in Q3 after making $974 million in Q2, and it has $895.4 billion in commitments, "substantially all of which is related to compute and infrastructure." Bloomberg reports backers now talk about a $1.2 trillion to $1.3 trillion valuation, down from $1.4 trillion to $1.5 trillion before the selloff, and The Information reports Anthropic is telling investors Claude helps design its chips, which knocked Synopsys and Cadence down as much as 4%. Read more

🔁 An outside evaluator spent 11 weeks inside Anthropic and estimates its AI progress has sped up 1.9x since October 2025. METR measured the speedup on Anthropic's internal benchmark of well-specified tasks and guesses "perhaps ≈80%" of it comes from AI doing the work, while noting that Anthropic "redacted a majority of the evidence we found most informative." The same week, Anthropic's amended S-1 said its expectations for future models "indicate that we may imminently meet" its own threshold for "dramatically accelerated progress attributable to AI." Read more

🔓 OpenAI's GPT-6 Cyber produced 948 working Linux exploits in one test campaign, and only vetted security teams can use it. It's OpenAI's second model rated "Critical" for cybersecurity after GPT-6 Astra, scoring 48.45% on the ExploitGym hacking benchmark to Astra's 42.43%, and it found 81 previously unknown high-severity Firefox bugs. Access is limited to customers in OpenAI's Daybreak Red program. Read more

🧪 AI Research of the Week

Just the Facts: How Dialogues with AI Reduce Conspiracy Beliefs
From Carnegie Mellon, Cornell and MIT

Jake's Take: In 2024 three researchers showed that arguing with GPT-4 for a few rounds made people believe their favorite conspiracy theory less. This new follow-up from the same researcher digs deeper into why that is.

They had 2,193 Americans each name a conspiracy theory they believed, rate it from 0 to 100, and then argue about it with an AI for three rounds:

• The standard conversation cut belief by 10.9 points with GPT-4 and 14.6 points with GPT-5.2 (a chat about cats vs. dogs moved it 1.4)

• Telling people up front that the AI would try to change their mind still got 9.1 points

• Shorter answers, a debate framing, or a cooler "just inform them" tone made no real difference

• Telling the AI to persuade without facts and evidence dropped it to 3.8 points

In the GPT-5.2 study, 23.1% of the people who started out believing ended on the "probably false" side, compared with 1.5% of the group that talked about something else.

Worth noting: everyone here volunteered for an online study, which means the most committed believers are probably missing from the study. Also, the "no facts" version stripped out facts and reasoning together, and nothing measured how long the change lasted. It’s probably hard to convince diehard conspiracy theorists to jump into a study like this, but I’d like to see them try next time.

what to know for later

💵 OpenAI closed the first $11 billion of its round at a $1.25 trillion valuation. Bloomberg and Reuters reported it on November 6, the day after the selloff. OpenAI first asked for $30 billion or more at $1.4 trillion, and the round could reach about $20 billion in the coming weeks. Read more