D.A.D. joke of the day: I asked ChatGPT for a joke about job security. It said, "I'm working on it."
What's New
AI developments from the last 24 hours
Nvidia Loses Its Founder, and AI Loses Its Chief Salesman
AI Business & Money Chips, Data Centers & Energy Nvidia
Jensen Huang, who co-founded Nvidia in 1993 and built it into the company at the center of the AI boom, resigned Thursday as chief executive, president and board member, effective immediately. He won't work with Nvidia "in any capacity going forward," a spokesperson said. Chief financial officer Colette Kress is interim CEO while keeping her finance job, and the board, which named Harvey Jones its first-ever chair, will search inside and outside the company for a permanent successor. Nvidia gave no reason. "There is no excuse for what I did," Huang wrote to staff.
His exit came a day after The Wall Street Journal reported the board was preparing to oust him, following his admitted affair with Alice Huang, an executive U.S. officials believe is a Chinese spy, and a second affair, reported by The New York Times, with an Nvidia executive he managed.
Investors were shaken but not panicked. Shares plunged more than 13% at Thursday's open, then recovered to close down 4.3%, erasing about $182 billion in market value, CNBC reported. Nvidia reports quarterly results next week. Analyst Dan Ives said the resignation "ends the dark chapter"; others faulted the board for never grooming a successor. President Trump called Huang "a legend" and urged Nvidia to hire Apple's Tim Cook, though nothing suggests Cook is a candidate.
The bigger loss may be what Huang did outside Nvidia. He was the AI industry's chief salesman to governments, its loudest champion of open models (D.A.D., July 25) and its most effective voice against limits on chip sales to China. Nvidia hasn't said whether its push into free, open AI models and agent-safety tools continues without him.
Why it matters: Nvidia's chips will keep powering the AI buildout; engineers, not one executive, make them. But Huang was the one person who could work Trump, Beijing and every major AI lab at once. His exit leaves Nvidia without its founder, its best lobbyist and its public face just as Congress moves to tighten chip exports, and it hands Washington more leverage over the world's most important AI company.
Sources: CNN · CNBC · NBC News · Axios · Fox Business · The Wall Street Journal
Google Launches Faster Gemini Model That Auto-Switches for Complex Tasks
Google released Gemini 4 Hydrogen, a lightweight model built for everyday coding tasks and AI agents that complete multi-step jobs on their own. It's now live across Google AI Studio, the Gemini API, the Gemini app, Search, and Antigravity. A new feature called Orchestrator automatically switches between Hydrogen and Google's larger models depending on task difficulty, matching top-tier performance while using less computing capacity and finishing faster, according to Google's internal benchmarks. It also substantially outperforms Google's prior fast model on coding and accuracy tests.
Why it matters: This follows Google's November 12 launch of a quick-decision model (D.A.D., November 12) and signals the industry is racing to build cheaper, faster models that can be trusted to run multi-step work without constant human checking.
Discuss on Hacker News · Source: blog.google
Anthropic Set to Go Public With Nasdaq Debut Next Week
Anthropic is going public. The company kicked off its IPO roadshow November 10th, with pricing expected November 18th and shares set to begin trading on Nasdaq under ticker 'ANT' the following day. Anthropic disclosed revenue of $17 billion, up more than 12-fold year over year, and over 11,900 customers paying at least $100,000 annually—also up 12-fold. The company posted a $23.3 billion net loss, though most of that reflects a one-time accounting charge tied to its funding history rather than cash burn. Investors had reportedly pushed for a lower IPO price after the recent AI stock selloff (D.A.D., November 7).
Why it matters: Anthropic would be the first of the leading U.S. AI labs to go public, meaning ordinary investors—and public markets broadly—will now get a real-time verdict on whether AI spending and growth can justify its costs.
Discuss on Hacker News · Source: ipo.anthropic.com
AI App Flood Reportedly Overwhelms Apple's Review Process, Developers Say
Apple's App Store is reportedly getting flooded with 10-20 new AI-generated music apps daily, and developers say the review process—apparently leaning on automation to cope—is buckling. Pittsburgh Modular's Microvolt 3901 saw its iOS release delayed weeks past its desktop launch; Unfiltered Audio's Sinusoidal Run Rhythm was rejected four times, including one reviewer allegedly inventing a licensing requirement for a made-up term, "Polyharmonic," and flagging a MIDI variant as spam. Apple promises 90% of reviews in under 48 hours; developers describe waits of weeks.
Why it matters: If AI flooding an app store can jam up the human (or AI) gatekeepers meant to police it, the same dynamic—content floods overwhelming review systems—is coming for any platform that relies on manual or semi-automated approval.
Discuss on Hacker News · Source: synthanatomy.com
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
After Huang, Washington Turns on the Chip Pipeline to China
Regulation & Policy China & Geopolitics Nvidia
Huang's resignation hasn't ended the scandal in Washington; it has widened it. The question now is how Nvidia's most advanced chips keep reaching Chinese AI labs, and who in the U.S. government knew the risks.
Congress is moving fast. House Speaker Mike Johnson said Thursday there is a "growing consensus" to limit chip exports to China, Reuters reported, and he backs adding restrictions to the year-end defense bill, as does House Democratic leader Hakeem Jeffries. Rep. John Moolenaar, the Republican who chairs the House China committee, went further: "American companies should never jeopardize our nation's security by giving Chinese companies access to their most advanced chips," he said, and Nvidia must "appoint leadership that makes national security its top priority." It's an extraordinary message from a Republican, since the Nvidia chips sold to Chinese companies were licensed by the Trump administration. Huang has agreed to testify.
Democrats are opening another front. The top Democrats on the Senate Finance and House Ways and Means committees launched an investigation into what they call the administration's "scheme to approve and tax exports" of Nvidia chips to China, a deal that gives the government a 25% cut, and told Commerce Secretary Howard Lutnick to preserve records. In January, they become the committees' chairs. Lutnick is under pressure on two fronts: Axios reported that he repeatedly stalled a crackdown Trump approved on companies suspected of smuggling Nvidia chips, which Commerce partly disputes, and The New York Times reported that a senior Commerce official gave Nvidia permission to contact investigators probing Megaspeed, Alice Huang's company. The White House says Trump has "complete confidence" in Lutnick.
So what is China doing? Two things at once. First, Beijing is blocking most official imports of the chip Washington approved for sale, Nvidia's older H200: about 600,000 are sitting in Hong Kong waiting for Beijing's go-ahead, Reuters reported last week. Beijing hasn't fully explained why, but reports have tied it to a push to build up Chinese chipmakers such as Huawei. Second, Chinese AI labs are still getting Nvidia's newest, most powerful chips, the ones the U.S. bans outright. They rent them at data centers outside China, a practice Reuters reported Thursday is accelerating, and, U.S. officials allege, some chips are smuggled in. That is the heart of the Huang scandal: Alice Huang helped run Megaspeed, a Singapore data-center company that bought $4.8 billion of Nvidia chips and was investigated for smuggling them to China. What isn't known is whether Beijing directs or merely tolerates these workarounds, or how many chips get through. A forensic analysis published Thursday by researcher Will DePue found signs in the labs' own published models that DeepSeek, Moonshot, MiniMax and Ant Group recently switched to Nvidia's top Blackwell chips, which can't be exported to China. It hasn't been independently reviewed, and it can't show whether the chips were in China or rented abroad. The bill that would close the rental loophole, the Remote Access Security Act, passed the House 369-22 in January and has sat in the Senate since.
Why it matters: The fight is shifting from whether to sell chips to China to whether Washington can stop them leaking out. Julian Gewirtz, a former senior director for China at the National Security Council, argues in The Atlantic that the real scandal is that officials "were warned about the danger and nevertheless continued relying on him." With weeks left before Democrats take control, this lame-duck Congress may act first.
Sources: Reuters, via Yahoo · House Select Committee on China · Senate Finance and House Ways and Means Democrats' letter · Axios · The New York Times · Reuters, via The Next Web · Unemployment Labs · Congress.gov · The Atlantic
What's in the Lab
New announcements from major AI labs
Students Outpace Teachers in Classroom AI Use, OpenAI Study Finds
OpenAI published findings from ThinkYoung, a European youth think tank, alongside lessons from its education partnerships across Europe, the Middle East and Africa. The piece argues schools face a widening "guidance gap"—students are adopting AI chatbots for schoolwork faster than teachers are getting trained to oversee that use—and calls for more structured AI literacy programs and teacher support rather than blanket bans or unmonitored access.
Why it matters: OpenAI has a commercial stake in classroom adoption, but the underlying problem—institutions lagging behind student AI use—is a real test of whether schools can adapt policy as fast as the technology spreads.
Singapore Airlines Deploys OpenAI Chatbots Across Customer Service, Trip Planning
Singapore Airlines says it's rolled out OpenAI's models and API across customer support, trip-planning, and internal staff tools. The airline claims customer satisfaction scores for its Kris chatbot doubled after the upgrade, though it hasn't disclosed the underlying numbers or methodology behind that figure. The deal, part of OpenAI's ongoing push into enterprise partnerships, adds a major airline to the growing list of travel and hospitality brands embedding chatbots directly into customer-facing operations rather than just back-office tasks.
Why it matters: Airlines have long had notoriously bad customer service bots, so a brand-name carrier publicly crediting AI for a satisfaction jump is a signal other travel companies will be watching closely, even without independent verification of the claim.
OpenAI Builds Internal Kill Switch to Rein In Its AI Models
OpenAI unveiled Control Room, an internal system that tracks model workloads running across its research infrastructure and lets engineers rapidly shut them down if safety or security issues emerge. The company frames it as operational infrastructure—a kill switch and monitoring layer for the computing jobs that train and test its models, rather than a consumer-facing feature. Details on how it was built or tested were not disclosed.
Why it matters: As AI labs race to build more autonomous and powerful systems, the ability to quickly see what's running and pull the plug becomes a basic safety requirement rather than an afterthought.
What's in Academe
New papers on AI and its effects from researchers
People Trust AI Advice More Than Humans, But It Doesn't Change Their Choices
A new study found that people rated AI chatbot advisors as more rational, trustworthy, and convincing than human advisors when discussing an actual charitable-donation decision with real money on the line. AI advisors also shared more facts and more readily offered opinions and justifications. But here's the catch: despite the warmer reception, participants' actual donation choices were no different whether they got advice from the AI, a human, or no one at all.
Why it matters: It suggests people may trust AI guidance more than it actually changes what they do—a gap worth watching as companies roll out AI advisors for financial, health, and workplace decisions.
AI Models Struggle to Predict Personality From Social Media Posts
A new study tested whether leading AI models—GPT-4, Gemma, Llama 3.1, and Mistral—could predict how a specific person would answer a standard personality test, using that person's actual social-media posts as context. The result: feeding models more of someone's writing didn't reliably improve accuracy over generic guesses. Models tended to default toward average personality scores and struggled at the extremes. GPT-4 did better when given an explicit personality profile directly, rather than inferring one from text, particularly for traits like extraversion.
Why it matters: As companies explore AI 'digital twins' for market research, HR screening, or personalized marketing, this suggests current models are better at applying stated traits than genuinely inferring someone's personality from how they write.
Caregivers Warn Home Monitoring AI Could Add Surveillance, Liability
A new study brought together 24 professional caregivers—18 personal health aides and 6 certified nursing assistants—for workshops exploring a hypothetical ambient sensing system designed to monitor elderly or disabled patients in their homes. The caregivers raised concerns largely absent from current AI design conversations: such systems could add to their workload, create new liability, and expose their work to constant surveillance by employers or family members, even as the technology promises to help patients age in place.
Why it matters: As ambient AI monitoring spreads into home care and assisted living, this research signals that the aides and nurses actually using these systems day to day may be designed around rather than designed with.
AI Image Tools Show Same Future City, Study Finds
A study of 907 images generated by Microsoft's Bing Create (built on DALL-E 3) found the tool produces a strikingly uniform vision of future cities—hovercrafts, rooftop gardens, solar panels in blue-green palettes—regardless of which of 12 countries or 15 languages prompted it. Researchers argue this flattens cultural difference into a single Western-style template, scrubbing out poverty, conflict, and local politics, and echoing older Orientalist stereotypes rather than reflecting how different places actually imagine their own futures.
Why it matters: As these tools get embedded in education, urban planning, and media, their narrow default imagery risks quietly homogenizing how people worldwide picture progress and development.
AI Stand-Ins Could Help Gather Public Input on AI Rules
Researchers tested whether AI 'agents' could stand in for individual citizens when deliberating on AI policy questions, letting companies or governments gather broader public input without polling everyone directly. In a study of 746 participants, these agent-represented opinions closed 68% of the satisfaction gap and 82% of the fairness gap compared with people weighing in themselves. Participants rated the approach fairer and more useful than funneling input through a single AI model, but most still preferred speaking for themselves when given the choice.
Why it matters: As AI labs and regulators face pressure to justify policy decisions as reflecting public input, this points to a cheaper middle ground between full public consultation and letting AI decide alone—though the finding that people still prefer direct participation suggests agents are a supplement, not a replacement.
What's On The Pod
Some new podcast episodes
AI in Business — How to Choose a Compute Platform Your Products Won't Outgrow - with Kyle Larson of IDEMIA