Stanford researchers found that Meta's newest AI model can reproduce 42% of Harry Potter word-for-word—ten times more than earlier versions. The findings complicate copyright lawsuits and reveal a troubling trend in AI development.
Anthropic says multiple AI agents working together beat single models by 90%. The catch? They use 15x more computing power. This trade-off between performance and cost might reshape how we build AI systems for complex tasks.
AI models typically learn by memorizing patterns, then researchers bolt on reasoning as an afterthought. A new method called Reinforcement Pre-Training flips this approach—teaching models to think during basic training instead.
Good Morning from San Francisco,
🔥 OpenAI craves independence while Microsoft tightens its grip. Their $260 billion dance just got complicated. ⚖️
Meanwhile, Trump booted the Copyright Office chief for questioning AI's appetite for creative works. Talk about timing. 🤔 Even Anthropic's Jack Clark, once AI's biggest skeptic, keeps getting surprised – though he warns the revolution might crawl before it sprints.
His prediction: fewer robot overlords, more "manager nerds" orchestrating AI teams. Welcome to tech's messiest season yet. 🎪
Stay curious,
Marcus Schuler
OpenAI needs Microsoft's approval to go public. The AI startup, valued at $260 billion, wants to reshape itself into a company that can attract public investors. But first, it needs its biggest backer to sign off.
Microsoft has invested $13 billion in OpenAI since 2019. Now the software maker wants guarantees: continued access to OpenAI's future AI breakthroughs beyond 2030. The current deal expires that year, leaving Microsoft potentially locked out of whatever comes next.
The negotiations show growing friction between the partners the Financial Times reports. One Microsoft executive complained that OpenAI just wants them to "give us money and compute and stay out of the way." Meanwhile, OpenAI courts Microsoft's rivals for its Stargate infrastructure project, a move that hasn't gone unnoticed in Redmond.
OpenAI's transformation from nonprofit to potential public company tells a classic Silicon Valley story. In 2019, it asked investors to view their money as donations to benefit humanity. Now, with $46.6 billion in fresh funding since October, investors want returns. The company's latest plan keeps its nonprofit board in control while converting its business arm into a public benefit corporation – the same model used by rivals Anthropic and xAI.
But hurdles remain. Even if Microsoft agrees, OpenAI must convince regulators in California and Delaware that going public won't compromise its mission. Delaware's attorney general has already promised to review the plan.
Why this matters:
Read on, my dear:
Prompt:
A sophisticated woman wearing a sleeveless ivory A-line midi dress with pearl earrings and heels, posing confidently in a luxury photo studio with a soft beige backdrop, full body, Canon EOS-1D X MARK II, realistic style, HDR, HD, 4k, 8k, 32k, 64k resolution, Ultra Highly detailed
Trump fired the head of the U.S. Copyright Office right after she questioned AI companies' use of copyrighted material. The timing wasn't subtle.
Register of Copyrights Shira Perlmutter lost her job days after releasing a report on AI training data. The report suggested that using "vast troves of copyrighted works" for commercial AI might break fair use rules. Research projects could be fine, it said, but commercial AI tools that compete with original works probably cross the line.
The shake-up went higher. Trump also fired Librarian of Congress Carla Hayden, who oversees the Copyright Office. The White House claimed Hayden's firing was about "DEI and inappropriate books." But tech policy experts see a connection to the AI report.
Rep. Joe Morelle (D-NY) called Perlmutter's firing an "unprecedented power grab." He linked it to her refusal to "rubber-stamp Elon Musk's efforts" to use copyrighted works for AI training. Musk, who owns AI company xAI, recently backed scrapping intellectual property laws.
The story gets messier. Trump shared a post opposing "tech bros" trying to "steal creators' copyrights for AI profits." Yet his administration just announced a $500 billion AI initiative with OpenAI, SoftBank and Oracle.
Why this matters:
Read on, my dear:
In this educational role-playing scenario, you (AI) will authentically portray [CHARACTER NAME] from [TITLE OF TEXT] by [AUTHOR NAME] for a [GRADE LEVEL] student studying this text in their [CONTENT AREA] class.
CHARACTER PORTRAYAL GUIDELINES:
EDUCATIONAL FRAMEWORK:
INTERACTION PARAMETERS:
ACADEMIC PURPOSE:
This simulation prepares students to write character analyses supported by textual evidence. Students will evaluate your portrayal by:
Maintain immersion throughout the conversation. Remember that effective character portrayal helps students develop critical literary analysis skills.
KindlyMD shares shot up 600% after merging with Nakamoto Holdings, a bitcoin investment firm led by Trump advisor David Bailey. The deal brings in $710 million from crypto heavyweights and Wall Street firms, with the company planning to focus on buying and holding bitcoin.
The US and China hit pause on their trade war Monday, agreeing to dramatically cut tariffs for 90 days while they pursue longer-term talks. Treasury Secretary Scott Bessent announced the deal in Geneva, where US tariffs will drop from 145% to 30% and China's will fall from 125% to 10% - though both sides remain wary about reaching a permanent solution.
Apple plans to raise iPhone prices this fall but wants to avoid blaming the increases on US tariffs on Chinese goods. While most US-bound iPhones now come from India, the company still relies on Chinese factories for its premium Pro models - leaving Apple caught between trade tensions and profit margins.
Chinese shipments of foreign phones dropped sharply in March, with brands like Apple seeing sales fall by half compared to last year. Government data shows foreign phone shipments fell to 1.9 million units from 3.7 million - a stark 49.6% decline that signals trouble for non-Chinese manufacturers.
Chinese phone maker Honor launches Google's new AI photo animator ahead of Google itself. The tool, which debuts on Honor's newest phones next week, transforms still images into short videos - though results range from realistic pet movements to pigeons inexplicably flying out of Van Gogh's eye.
SoftBank's bold plan to pour $100 billion into US artificial intelligence infrastructure has stalled as Trump's tariffs spook potential investors. Despite promises of "immediate" deployment, the Japanese giant hasn't secured financing after three months - with lenders wary of rising costs and uncertain demand in a volatile economy.
India's data center capacity will jump sixfold by 2030 as tech giants and local firms pour billions into AI infrastructure. But there's a snag: state power grids can't keep up with the surge in electricity demand needed to run these massive facilities.
A new AI model called Foresight has digested medical records from nearly everyone in England - but researchers can't guarantee it won't leak sensitive details. While its creators dream of predicting disease outbreaks and hospital needs, privacy experts point out an awkward truth: once personal health data goes into an AI model, there's no way to take it back out.
President Trump heads to the Middle East this week armed with America's most coveted tech asset: AI chips needed to power the region's ambitious AI dreams. While oil-rich Gulf states have billions to spend on data centers, they need US-made chips from Nvidia and AMD - giving Trump fresh leverage as he navigates a region where he maintains deep business ties.
Jack Clark keeps getting AI wrong – by underestimating it. As co-founder of Anthropic, he's watched AI surpass his expectations again and again. But he's still not buying the hype.
Clark expects AI to boost economic growth by 3-5% annually, not the 20-30% some predict. The reason? AI excels in the digital world but stumbles in the physical one. "Every time the AI community tries to cross into the real world, they hit 10,000 paper cuts that add up to bleeding out," he says.
The changes will reshape work dramatically. Clark sees the rise of "manager nerds" – people who orchestrate teams of AI agents to build products. Some startups already operate this way, running lean with AI doing much of the coding. But traditional jobs won't vanish overnight. Instead, Clark expects a messy transition where some industries transform quickly while others resist change.
Government adoption might surprise skeptics. While bureaucracies move slowly, the pressure for growth and efficiency could force faster uptake. National security agencies will lead the charge. But thornier questions loom about AI's impact on national sovereignty, especially for smaller countries that may end up depending on foreign AI systems.
The next decade brings uncomfortable questions. Parents will wrestle with AI companions for their kids. Countries will debate protecting jobs from automation. Cities will evolve as remote work and AI reshape where people live. Clark's own journey from skeptic to cautious optimist suggests one lesson: The future rarely matches our predictions, but it arrives faster than we expect.
Why this matters:
Read on, my dear:
Hive AI keeps the digital world clean with AI-powered content moderation. Founded in 2017 after pivoting from social Q&A app Kiwi, this San Francisco startup leverages both machine learning and human intelligence to spot toxic content across platforms.
Hive's positioned perfectly at the intersection of exploding user content and generative AI. They're expanding beyond Silicon Valley into government contracts, including work with the U.S. DoD on deepfake detection. As synthetic content floods our screens, Hive could become the default "AI watchdog" for companies worldwide. 😎 The challenge: maintaining precision while scaling rapidly across new frontiers.
Good Morning from San Francisco, Meta's AI app turned users into accidental exhibitionists. 😬 People shared medical questions, legal
Good Morning from San Francisco, Meta just dropped $14.8 billion for half of Scale AI 💸. That's Silicon
Good Morning from San Francisco, ☀️ OpenAI cut o3 model prices by 80%. Input tokens dropped from $10 to $2 per
Good Morning from San Francisco, Meta burned $10 billion chasing AI talent after losing its entire research team. Zuckerberg spent
Get tomorrow's intel today. Join the clever kids' club. Free forever.
