DeepMind's newest AI system has found ways to speed up calculations that stumped mathematicians since 1969. The tool is already saving money in Google's data centers - and might reshape how we write computer code. Its first real-world results point to an unexpected shift in algorithm design.
OpenAI just added a specialized coding model to ChatGPT, but that's not the whole story. The company also opened up about its safety testing - a move that could reshape how AI firms handle transparency. Here's what changed and why it matters.
Silicon Valley, who once begged for AI guardrails, now wants them gone. Their new battle cry? Beat China, worry later. 🏃♂️
Sam Altman led the flip-flop parade at Capitol Hill. The OpenAI chief, previously a regulation fan, now warns of "disaster" if government slows AI development. Microsoft and AMD joined the chorus, demanding faster permits and more power - literally. Those AI data centers are hungry beasts. 🔌
Meanwhile, using AI at work? Watch your back. 🤖 Duke researchers found it marks you as lazy and incompetent - unless your boss uses AI too. Talk about artificial anxiety.
Stay curious,
Marcus Schuler
Silicon Valley Wants Lighter AI Rules to Beat China
Tech executives who once championed AI regulation now warn it could derail American leadership in the field. The shift marks a stark departure from the cautionary tone that dominated discussions just two years ago.
OpenAI CEO Sam Altman, who previously backed strict oversight, told senators Thursday that requiring government approval for AI releases would be "disastrous." Microsoft's Brad Smith echoed the call for a "light touch" approach, while AMD chief Lisa Su stressed the need to run "faster" in global competition.
The hearing highlighted how power supply could make or break US AI ambitions. Tech leaders described a pressing need for more data centers, faster permitting, and abundant electricity - with projects like OpenAI's $500 billion Stargate facility in Texas showing the massive scale required.
China's AI progress, exemplified by DeepSeek's recent advances, looms large over policy discussions. While US companies maintain a lead in key areas, executives warn that overly restrictive rules could hand advantages to Chinese competitors.
The testimony comes as President Trump moves to dismantle Biden-era AI regulations, including complex rules on chip exports. Industry leaders backed the change while acknowledging some controls remain necessary for national security.
Why this matters:
The race for AI dominance increasingly resembles the 1990s internet boom - but with much higher stakes for national security and economic power
Unlike social media's rise, AI development requires massive physical infrastructure, turning mundane issues like power grid capacity and construction permits into strategic priorities
Prompt: Beautiful Korean lady holding shark, looking, bluebackdrop, waterbackdrop, watersplashes, 3D, architecture, diamond necklaces, diamond liprings, diamond noserings, diamond lipstick, next to shark, 32KUHD photography, biopunk, large big strong lady, 3D, sculptural, surrealism, fashion luxury, watches, necklaces, pendants, big hair, messy hair, blue hair, white hair, multicolored hair, closeup 70mm, magazine photo
Self-Teaching AI Matches Human-Trained Systems
A research team has built an AI that teaches itself without copying humans. The system, called Absolute Zero, creates and solves its own practice problems in coding and math - and it's beating traditional AI models at their own game.
The breakthrough comes from an unlikely alliance: researchers at Tsinghua University, Beijing Institute for General AI, and Penn State worked together to bypass a core limitation of current AI systems - their reliance on human-created training data.
Absolute Zero works like a student who writes their own practice tests. It generates programming exercises and math problems, then learns from both right and wrong answers. In testing, it matched or beat specialized AI systems trained on thousands of human examples.
The system proved particularly strong at transferring skills between different types of problems. When tested on new coding challenges, it outperformed AI models specifically built for programming tasks.
The approach worked across various AI model sizes, with bigger models showing more dramatic improvements. This suggests the method could become more powerful as computing capacity grows.
Major AI labs have struggled to find enough quality training data as their systems grow more sophisticated. Absolute Zero's self-teaching approach might solve this bottleneck - if it can scale beyond coding and math.
Why this matters:
AI might soon learn advanced skills without needing human examples first, removing a key barrier to progress
The success in both coding and math hints that self-directed learning could work for other complex reasoning tasks
California Cuts Back AI Privacy Rules After Pressure
California's privacy watchdog just backed down from tough AI rules after pressure from Governor Newsom and tech companies, slashing compliance costs from $834 million to $143 million and freeing 90% of businesses from oversight. The retreat means companies like Google and Meta won't need risk assessments for their targeted advertising, while employers can dodge scrutiny by claiming their AI tools are merely "advisory."
Palantir Joins Top 10 US Tech Firms with Sky-High Valuation
Palantir's stock price surged 8% Thursday to hit $281 billion in value, pushing past Salesforce into the top 10 US tech companies. But there's a catch - it sells just one-tenth of what Salesforce does and trades at 520 times its earnings.
AI Makes Fraud 4,000 Times Easier, Says 'Catch Me' Expert
Frank Abagnale, the con artist who inspired "Catch Me If You Can," says AI and social engineering let fraudsters steal millions without ever meeting their victims. While criminals exploit these tools, banks and businesses are turning to quantum AI to detect scams in minutes instead of months.
Senate Blocks Crypto Bill After Trump Memecoin Controversy
Senate Democrats blocked new stablecoin rules after Trump started promoting a cryptocurrency bearing his name. The vote failed 49-48, with Democrats demanding a ban on presidents profiting from crypto while in office. Trump had advertised a memecoin offering dinner seats to its biggest buyers.
Chrome Gets Gemini-Powered Scam Blocker
Google just added Gemini Nano to Chrome to spot scams before they happen. The AI runs on your device to flag suspicious sites instantly, while new Android warnings help users dodge spammy notifications.
Bill Gates Plans to Spend $200 Billion, Then Close Foundation
Bill Gates will spend his remaining fortune and shut down his foundation by 2045, betting that new AI tools can accelerate global health progress despite Trump's foreign aid cuts. The plan would pour $200 billion into fighting diseases and poverty over 20 years, aimed at goals like eliminating malaria in Africa and halving childhood deaths.
New Headphones Let You Understand Multiple Languages at Once
Researchers built AI headphones that translate group conversations in real time, keeping each speaker's unique voice and location. The system works with regular noise-canceling headphones plugged into a laptop, turning French, German or Spanish into English with just a few seconds delay.
Cambridge Team Beats Supercomputers at Weather Prediction
Cambridge researchers built an AI system that predicts weather faster and better than giant supercomputers - using just a laptop. Named Aardvark Weather, it could help vulnerable countries spot deadly storms and heat waves without needing expensive technology.
Study: Using AI Tools at Work Carries Social Cost
Using AI at work improves productivity but hurts your reputation, new research shows. A Duke University study found that employees who use AI tools face harsh judgments from colleagues who see them as lazy and less competent.
The research team ran four experiments with over 4,400 participants. They found that workers who use AI expect to be judged negatively - and they're right. Managers and coworkers consistently rated AI users as less diligent and capable than people who use traditional tools or get help from other sources.
The stigma cuts across age groups, gender, and job types. But two factors can soften the blow: when AI is clearly useful for the task, and when the person doing the judging uses AI themselves. Managers who don't use AI were less likely to hire AI users, while those familiar with the technology showed less bias.
These findings create a tough choice for workers. AI tools can make them better at their jobs, but using them might hurt their professional image. Many employees respond by hiding their AI use - a strategy that could slow adoption of helpful technology.
"This social penalty reminds us of past resistance to new tools like calculators," says lead author Jessica Reif. "People worried that using calculators would prevent students from developing real math skills. We're seeing similar fears about AI."
Why this matters:
The research reveals an overlooked barrier to AI adoption: social pressure rather than technical limits
As with past technologies, the stigma may fade as AI becomes normal in workplaces
I need a system for analyzing stock market trends. Please include:
A detailed explanation of how to use moving averages (specifically SMA, EMA, and MACD) to identify trend directions and potential reversals
Guidelines for setting appropriate timeframes based on my trading style (day trading vs. swing trading)
Three additional technical indicators that complement moving averages, with specific entry/exit rules
A step-by-step process for screening stocks that fit these criteria
Common pitfalls to avoid when using these indicators
A simple example analyzing a real stock using this system
I have intermediate trading knowledge but need practical implementation steps rather than just theory.
🚀 AI Profiles: The Companies Defining Tomorrow
Mistral AI: Europe's Leading AI Startup
Paris-based Mistral AI emerged in 2023 as Europe's answer to OpenAI, building frontier language models with a distinctly European approach to AI independence. The French startup has rocketed to unicorn status on a mission to create world-class AI while championing openness and technological sovereignty.
👨💻 The Founders
Founded in February 2023 by three AI virtuosos: CEO Arthur Mensch (ex-DeepMind), CTO Timothée Lacroix and CSO Guillaume Lample (both from Meta's AI research)
All graduates of France's elite institutions (École Polytechnique, École Normale Supérieure)
Started with ~50 employees, now scaled to several hundred based in Paris
Created to fill Europe's gap in frontier AI development and reduce dependence on US tech giants
🚀 The Product
Suite of large language models: Mistral Large 2 (flagship text model), Pixtral (multimodal for images), Codestral (coding specialist), and "Les Ministraux" (mobile-optimized compact models)
"Le Chat" - conversational AI assistant with web search, image generation, document analysis, and code interpretation capabilities
Differentiates through open-source approach, releasing several models under permissive licenses while keeping premium versions for commercial use
Offers enterprise deployments with self-hosting options prioritizing data sovereignty and privacy
🥊 The Competition
Battles tech giants like OpenAI (backed by Microsoft), Google (DeepMind), Meta, and Anthropic in the global AI race
Leads the European AI startup ecosystem, outpacing Germany's Aleph Alpha and establishing continental dominance
Leverages European identity as advantage for privacy-conscious clients and government contracts
Enjoys political support from French leadership as the continent's AI champion
💰 Financing
Raised €1 billion ($1.1B) in just two years across record-breaking rounds
Latest valuation: €5.8 billion ($6.2B) after June 2024 Series B led by General Catalyst
Investors include tech luminaries (Eric Schmidt), VC giants (Andreessen Horowitz, Lightspeed), corporate players (Microsoft, Nvidia, IBM), and French public funds
Revenue still modest (estimated $30M in 2024) but growing rapidly as enterprise adoption increases
🔮 The Future ⭐⭐⭐⭐
Mistral faces the classic startup challenge: turning technical brilliance into commercial success. The company must dramatically scale revenue to justify its unicorn valuation while competing against giants with deeper pockets. 🌪️ Yet its European identity, open approach, and growing government partnerships create genuine differentiation. IPO prospects look bright if Mistral can maintain its innovation pace and convert its early momentum into lasting market position.
