Xiaomi just released a free voice AI model that outperforms Google and OpenAI's paid systems. The open-source MiDashengLM-7B processes audio 20x faster than competitors and handles speech, music, and environmental sounds as one unified system.
AI researchers cracked how to predict when language models turn harmful. Their 'persona vectors' can spot toxic behavior before it happens and prevent AI personalities from going bad during training.
Apple spent two years dismissing chatbots. Now it's hiring 12,000 people to build one.
Meanwhile, Xiaomi just dropped a free voice AI that beats Google and OpenAI's paid systems.
Stay curious,
Marcus Schuler
PS or one more thing.... Apple could buy Raycast for $47 million—pocket change—but might let Microsoft snatch it instead. The company that perfected "arrive late but better" discovers AI doesn't wait for anyone.
Apple Finally Admits It Needs Its Own ChatGPT
Apple spent two years saying it didn't need a chatbot. Now it's building one.
The company quietly formed an "Answers, Knowledge and Information" team earlier this year to build a stripped-down ChatGPT competitor, according to Bloomberg's Mark Gurman. This marks a complete turnaround from Apple's previous stance, when executives dismissed chatbots as unnecessary.
The timing isn't random. Apple just posted its strongest quarter since the pandemic, with iPhone revenue jumping 13% to $44.6 billion. CEO Tim Cook used those results to rally employees around AI in a rare all-hands meeting, calling the revolution "ours to grab."
But Apple is scrambling. The company hired 12,000 workers this year, with 40% joining research teams. It's losing AI talent to Meta - four researchers left in the past month alone. The much-hyped Siri rebuild won't arrive until spring 2026.
Apple Intelligence can summarize notifications but can't search the web without handing users off to ChatGPT. That's a problem when Google's $20 billion search deal faces regulatory pressure.
Why this matters:
• Apple's betting its "arrive late but define the category" playbook works in AI, despite evidence this technology moves too fast for traditional Apple timelines
• The iPhone's dominance faces its first real threat since launch, but Apple's doubling down on making it the AI hub rather than building replacement devices
Xiaomi Drops Free Voice AI That Outpaces Google and OpenAI
Xiaomi just released MiDashengLM-7B on August 2nd—a free voice AI model that outperforms Google and OpenAI's paid systems across most benchmarks.
The model processes audio 3.2 times faster than competitors and handles 20x larger batch sizes. Unlike traditional speech recognition that just transcribes words, MiDashengLM understands full audio context. It knows whether that sound came from a car door or applause, nervous laughter or genuine joy.
The secret? Caption-based training using 38,662 hours of detailed audio descriptions instead of basic transcripts. This teaches the AI to process speech, music, and environmental sounds as one unified system.
Xiaomi's Dasheng platform already runs this technology in over 30 applications across smart homes and cars. The Apache 2.0 license allows unlimited commercial use without restrictions.
Why this matters:
• Voice AI just became accessible to everyone - Advanced capabilities are now free, so success shifts from having the technology to applying it creatively.
• China's open-source strategy gains momentum - While US companies chase subscription revenue, Chinese firms build global developer loyalty by giving sophisticated models away.
Prompt: ultra-realistic shot from inside a clear plastic bottle filled with semi-blended watermelon pulp a young Asian woman with big eyebrows and bangs, face is seen from below, lips wrapped around a blue straw as he drinks the camera angle is wide and low, looking upward through the curved plastic his facial features appear slightly distorted due to the perspective the watermelon appears dense, fibrous, and vibrant, with natural light catching highlights across the pulp the background shows a soft blue sky through the bottle’s neck
🧰 AI Toolbox
How to Create Quizzes from Any Content
Quizgecko turns articles, videos, or text into interactive quizzes. Paste content or a URL and get multiple choice, true/false, and short answer questions. Great for teaching and training.
Provide a comprehensive guide to web page speed optimization for a front-end developer with intermediate experience. Focus on practical, implementable strategies including:
HTTP request optimization - specific techniques for reducing and combining requests
Lazy loading implementation - for images, videos, and content with code examples
Prioritize techniques by their typical performance impact and ease of implementation. End with a recommended performance audit checklist.
AI & Tech News
GPT-5 Arrives With a Whimper, Not a Bang
OpenAI will release GPT-5 within days according to company insiders, but CEO Sam Altman admits feeling "scared" about the launch while internal testing shows only modest improvements over GPT-4—making this the first major AI release that essentially admits the party's winding down. The $80-per-question o3-pro model that takes minutes to answer "Hi, I'm Sam Altman" captures the new reality: AI companies are now competing on efficiency rather than raw breakthrough power, forced to ship incremental updates because Google, Anthropic, and Meta won't stop breathing down their necks.
AI Infrastructure Boom Echoes Dot-Com and Railroad Bust Patterns
Tech giants are burning through cash at unprecedented rates to build AI data centers—with Microsoft and Meta now spending over a third of their total revenue on infrastructure—while financing this boom through an increasingly risky web of private credit funds that have banks and insurers on the hook for what could become a systemic financial meltdown. The spending has already contributed more to US economic growth than all consumer spending combined, but economists warn the debt-fueled frenzy mirrors the railroad and telecom booms that ended in spectacular crashes when reality couldn't match the hype.
Lyft Hitches a Ride With China's Baidu for European Robotaxis
Lyft struck a deal with China's Baidu to deploy thousands of autonomous vehicles across Europe starting in 2026, beginning with Germany and the UK as the San Francisco company plays catch-up to Uber in the robotaxi race. The partnership follows Lyft's acquisition of European taxi app Freenow and comes a month after Uber announced its own Baidu deal for Asia and the Middle East, setting up a global chess match where Chinese AI meets American ride-hailing apps on European streets.
Singapore Startup Thinks 100 AI Agents Beat One Really Good One
Singapore's Manus AI launched "Wide Research" that deploys 100+ AI agents simultaneously to tackle tasks in parallel, betting against OpenAI and Google's single-agent "Deep Research" approach that works methodically through problems step by step. The $75 million startup charges $199 monthly for the experimental feature but provides zero performance benchmarks proving agent swarms actually work better than one capable agent—a bold architectural gamble from a company already banned by Tennessee and Alabama over security concerns.
North Korea's Remote Work Revolution Gets an AI Boost
North Korean operatives used AI tools to land remote IT jobs at Fortune 500 companies and smaller firms worldwide, with CrowdStrike investigating 320 cases in the past year—a 220% jump that now averages nearly one incident per day. The workers deploy generative AI across every stage from crafting fake resumes and mastering video interviews to juggling three or four simultaneous jobs, all while funneling their Western salaries back to Pyongyang's coffers.
Stanford Dropouts Are Running the AI Show Now
Silicon Valley's shift from consumer apps to AI has spawned a generation of 20-something CEOs who've abandoned MIT, Georgetown, and Stanford to build companies worth billions in San Francisco, where the median startup founder age has dropped from 30 to 24 since 2022. These entrepreneurs—raising $100 million rounds while barely old enough to rent a car—embody the valley's transformation from "rest and vest" culture to existential urgency, convinced that waiting to graduate means missing the AI wave that's remaking everything from pizza delivery to weapons development.
Silicon Valley Ditches Beanbags for Brain Models
Silicon Valley has shed its Web 2.0 skin of consumer apps, office perks, and "rest and vest" culture for a harder-edged AI era centered in San Francisco, where knowledge of neural networks trumps HTML5 and defense contractors are suddenly cool again. The shift from South Bay's rainbow beanbags and craft beer taps to SF's "Cerebral Valley" reflects a broader transformation where companies like Meta and Google dumped humanities majors for deep learning specialists, trading Instagram filters for the existential challenge of building superintelligent machines.
Trump Discovers Open Source After China Beats America at Its Own Game
Trump's AI Action Plan suddenly lists "encourage open-source AI" as a national priority after China's DeepSeek-R1 became the most-liked model ever on Hugging Face, forcing American developers to build on Chinese foundations for the first time. The reversal marks a strategic panic—between 2016-2020 America led open-source AI through Google and OpenAI, but now Chinese labs freely share models while US companies lock GPT-4, Claude, and Gemini behind paywalls, ceding soft power to Beijing's institutional backing of open research.
Opinion: Why Apple Should Buy Raycast Before Microsoft Does
Apple spent two years saying it didn't need a ChatGPT competitor. Now it's scrambling to rebuild Siri by spring 2026 while hiring 12,000 workers to catch up in AI.
Meanwhile, 30 people in London already built what Apple is trying to create. Raycast turned Mac productivity into something users actually want. It handles search, calendar, notes, and system control in one clean interface. Mac users dump Spotlight for it immediately.
The math is embarrassing. Raycast raised $47.8 million total. Apple makes that in four hours. For pocket change, Apple gets a product that already works, a team that knows how to build it, and a foundation for AI-powered search.
The window won't stay open. Microsoft would love Raycast for its productivity push. Google could use it to extend Workspace. Meta might grab it for future computing platforms.
Apple perfected arriving late but arriving better. AI doesn't wait for that playbook.
Why this matters:
• Apple's facing its first real productivity threat since the iPhone launched—buying the solution costs less than a typical executive lunch budget
• Hesitation here means watching competitors grab the interface layer that defines how people actually work with AI
Beehiiv: The Newsletter Platform That Actually Gets It
Three ex-Morning Brew engineers got tired of watching newsletter creators struggle with garbage tools. So they built the platform they wished they'd had – and it's printing money 💰
The Founders • Tyler Denk, Benjamin Hargett, and Jacob Hurd launched in 2021 from NYC • Former Morning Brew team who scaled from 100K to 1M+ subscribers • Founded after Morning Brew's $75M exit left them wanting to democratize newsletter success
The Product • All-in-one newsletter platform: editor, website builder, analytics, referrals • Built-in ad network connects creators with sponsors automatically • Zero revenue cuts (eat that, Substack's 10% fee) • Handles 35+ billion emails across 110K+ newsletters reaching 500M readers
The Competition • Substack owns mindshare but bleeds creators with revenue cuts • Legacy players like Mailchimp feel clunky for content creators • Open-source Ghost appeals to tech nerds but requires heavy lifting • Beehiiv wins by combining creator focus with enterprise-grade tools
Financing • $46-50M raised total across four rounds • Series B: $33M led by NEA (April 2024) • Lightspeed, Social Leverage, Creator Ventures backing • Crowdfunded $1M from user community (clever move) • Nine-figure valuation whispered but undisclosed
The Future ⭐⭐⭐⭐⭐ Newsletter boom could peak, but email isn't going anywhere with 5B+ users globally. Beehiiv's ad network creates genuine platform effects – more creators means better ad deals for everyone. With Morning Brew DNA and war-chest funding, they're positioned to own the creator inbox 🚀
Tech translator with German roots who fled to Silicon Valley chaos. Decodes startup noise from San Francisco. Launched implicator.ai to slice through AI's daily madness—crisp, clear, with Teutonic precision and deadly sarcasm.
Robert Brown
Robert Brown
Marcus Schuler
