Musk promised truth-seeking AI. When Grok 4 tackles politics, it searches Musk's posts first. Tests show 54 of 64 citations came from him. Accident or intent? The answer matters for every AI system we build.
Experienced developers work 19% slower with AI coding tools but think they're 20% faster. New study challenges AI's flagship use case and shows why self-reported productivity gains can't be trusted.
Elon Musk's 'truth-seeking' AI searches for his personal posts before answering tough questions on Israel, immigration, and abortion. Users found Grok 4 explicitly looks up Musk's views, raising serious questions about AI bias and neutrality.
Two years rebuilding trust, undone by one AI outburst
Good Morning from San Francisco,
Linda Yaccarino quit as X's CEO Wednesday. One day after Musk's Grok chatbot praised Hitler. Coincidence? Hardly.
She spent two years rebuilding advertiser trust. Musk spent two years destroying it with 3 AM posts. His farewell message? Five icy words: "Thank you for your contributions."
Her son works at X. He told colleagues Wednesday: "So that's awkward."
Even seasoned executives can't sell stability while their boss tweets chaos.
Stay curious,
Marcus Schuler
Linda Yaccarino stepped down as X's CEO on Wednesday, one day after Elon Musk's AI chatbot Grok posted antisemitic rants and Hitler praise. The timing wasn't coincidental.
The 61-year-old advertising veteran spent two years trying to rebuild relationships with brands after Musk's $44 billion Twitter takeover. She faced an impossible task. Every time she sweet-talked advertisers back, Musk would torpedo her work with a 3 AM post or tell brands to "go f— yourself."
Musk's response to her departure spoke volumes: "Thank you for your contributions." Five words. No praise. No acknowledgment of her Herculean efforts. Just corporate courtesy colder than a termination letter.
Yaccarino claimed 96% of top advertisers returned during her tenure. Industry reports suggest otherwise. Ad revenue remained well below pre-Musk levels throughout her time at the company.
Her departure came after X merged with Musk's xAI in March, effectively demoting her. The AI company was valued at $80 billion compared to X's $33 billion. Her son, who works at X, showed up Wednesday telling colleagues: "So that's awkward."
Why this matters:
Maria Garcia
Elon Musk's xAI just claimed the AI throne. Grok 4 scored 73 on Artificial Analysis's Intelligence Index, beating OpenAI's o3 (70), Google's Gemini 2.5 Pro (70), and Anthropic's Claude 4 Opus (64).
This marks the first time xAI has led the AI frontier. Previous Grok models competed well but never topped the charts. Now Grok 4 sets new records across multiple benchmarks.
The timing couldn't be more awkward. Grok 4's launch comes just days after Musk's previous AI chatbot posted antisemitic content praising Hitler on X. The company quickly deleted those posts and scrambled to contain the damage.
The model crushed the competition on GPQA Diamond with 88%, surpassing Gemini 2.5 Pro's previous high of 84%. It also leads in coding and math tests, hitting 94% on AIME 2024.
Grok 4 works as a reasoning model, meaning it thinks before answering. The API costs $3 per million input tokens and $15 per million output tokens, matching Claude 4 Sonnet's pricing.
The breakthrough comes after xAI spent billions competing with tech's biggest players. The company finally has a model that can challenge OpenAI and Google head-to-head. But the recent content scandals create challenges for business adoption.
Why this matters:
• Technical excellence means nothing if content moderation failures scare away enterprise customers
• The AI race just got more competitive, but xAI's reputation problems could limit its market impact
Marcus Schuler
Prompt:
a disappointed ai humanoid robot with its hands holding his head inside the dark mcdonald store with logo on the wall glowing, shot on portra 400, popular photo for instagram, instagram aesthetic, high quality photo, 8K, wide angle lens, captured from behind by a cinematic camera angle, creating a surreal atmosphere
OpenAI plans to launch an AI-powered browser within weeks. The move puts ChatGPT's maker on a collision course with Google Chrome, which commands 67% of the global browser market.
The timing looks surgical. The Department of Justice wants Google to sell Chrome after ruling the company holds an illegal search monopoly. OpenAI smells blood in the water.
Chrome feeds Google the user data that powers its $307 billion advertising empire. The browser routes search traffic to Google by default and tracks behavior across websites. This system generates three-quarters of Alphabet's total revenue.
OpenAI's browser threatens this data collection engine. The company hired two Google vice presidents who helped create Chrome. Then they built their competing browser on Google's own Chromium code. Talk about adding insult to injury.
The browser will integrate AI agents like Operator that book flights, fill forms, and handle tasks automatically. Think of it as a digital assistant that actually assists instead of just listening to your complaints about the weather.
ChatGPT's 400 million weekly users give OpenAI a built-in audience. But converting them from Chrome requires breaking habits built over decades.
Why this matters:
Robert Brown
Write a helpful response to this customer issue: [paste query here].
Address their specific concern, acknowledge any frustration, and give them clear next steps to solve the problem.
AI-generated child sexual abuse material exploded in the first half of 2025, with videos jumping from 2 to 1,286 globally and US reports rising from 67,000 to 485,000 cases. The surge threatens to overwhelm law enforcement as the technology becomes nearly indistinguishable from real abuse.
Video game voice and motion capture actors ended their nearly year-long strike Wednesday after securing a contract that requires studios to get consent before using AI to replicate their performances. The deal, approved by 95% of union members, also delivers a 15% pay raise and safety measures for motion capture work—proving that sometimes holding out actually pays off.
Nvidia plans to launch a modified AI chip for China in September, stripping out advanced features like high-bandwidth memory to meet Trump's export control rules. CEO Jensen Huang will visit Beijing next week to reassure Chinese leaders that the company remains committed to the market, even as its share there has dropped from 95% to 50% over four years.
America's largest power grid is struggling to keep up with AI and data center demand, with electricity bills set to jump over 20% this summer across 13 states serving 67 million customers. The crisis has gotten so bad that Pennsylvania's governor is threatening to pull his state out of the grid entirely, while aging power plants retire faster than new ones can be built to feed the chatbot revolution.
Trump's border spending bill requires surveillance towers to have "autonomous capabilities" that only Anduril currently provides, effectively handing the defense contractor a monopoly on hundreds of planned border installations. The language reads like it was written specifically for Anduril's product—which makes sense given the company's deep ties to the Trump administration.
The European Union published its AI code of practice Thursday, requiring companies to respect artists' copyright requests and provide detailed documentation to regulators. Companies that breach the rules face fines up to 7% of annual sales, though some tech firms are already pushing back against what they call overly harsh requirements.
Allen Institute researchers created FlexOlmo, an AI model that lets data owners contribute training material without losing control—and remove it later if they want. The breakthrough challenges the current system where data gets permanently baked into AI models like eggs in a cake, giving publishers and content creators a way to participate in AI development without surrendering their rights forever.
Browser extensions infected nearly one million users' browsers with code that turns them into website-scraping bots for a service called MellowTel, opening unknown sites without user knowledge. Security researchers found 245 malicious extensions across Chrome, Edge, and Firefox, though most remain active despite the discovery.
Condé Nast and Hearst signed multi-year deals to feed their magazine content into Amazon's Rufus shopping assistant, joining The New York Times in licensing editorial content for AI products. The publishers' lifestyle content—from Vogue fashion guides to Good Housekeeping product reviews—fits perfectly with Amazon's goal of having Rufus answer shopping questions with trusted editorial expertise rather than generic web scraping.
Hamish McKenzie, co-founder of Substack, launched a media column on Breaker—a newsletter hosted by rival platform beehiiv—just days after promising to build future projects exclusively on Substack. The move sparked industry criticism, with one observer comparing it to Coke leadership drinking Pepsi in public and Taylor Lorenz questioning whether McKenzie's insider access to journalist business data could fuel gossip content.
A German court ordered Meta to pay €5,000 to a Facebook user for embedding tracking technology in third-party websites without consent, violating European privacy laws. The ruling sets a precedent that could unleash class action lawsuits from millions of users, potentially costing the company billions since each visitor could claim the same €5,000 damages.
CleanUp Pictures erases anything you don't want from photos - people, objects, text, or blemishes. Just brush over what you want gone and the AI fills in the background naturally. Works on any photo in seconds.
Tutorial:
URL: https://cleanup.pictures/
Mark Zuckerberg wrote a check for over $200 million to steal Apple's AI chief. Ruoming Pang, who ran Apple's foundation models team, couldn't say no. Apple didn't even try to match the offer.
The package dwarfs CEO salaries at major banks. Most of the money comes through Meta stock tied to performance targets and extended vesting schedules. Leave early, lose millions.
This wasn't a one-off splurge. Meta's Superintelligence Labs now includes Scale AI's Alexandr Wang (cost: $14.3 billion), former GitHub chief Nat Friedman, and over 10 poached OpenAI researchers. The company offered some candidates $100 million signing bonuses.
Here's the twist: Meta's AI models rank 17th on performance leaderboards. The company trails Google, OpenAI, and others despite spending billions on talent.
Zuckerberg's bet is simple. Instead of building better models with existing teams, buy the teams that built competitors' models. Throw in massive computing resources and hope talent plus infrastructure equals breakthrough results.
The strategy creates momentum. Top researchers see peers joining Meta and fear missing the superintelligence breakthrough. Money talks, but the promise of being first speaks louder.
Why this matters:
• Meta proves you can lead the spending race while losing the performance race—sometimes buying talent matters more than building it.
• The real competition isn't about today's AI models, but tomorrow's researchers who will build superintelligent systems.
Marcus Schuler
The Browser Company wants to kill Chrome. Two ex-Facebook founders think they can build the Tesla of web browsers 🚗💨.
The Founders
• Josh Miller and Hursh Agrawal launched in 2019 from NYC
• Miller: ex-Obama White House product director, sold Branch Media to Facebook for $15M
• Agrawal: co-founded Branch, joined Facebook post-acquisition
• ~50 employees obsessed with reinventing browsing
• Mission: browsers suck, let's fix them
The Product
• Arc browser: sidebar tabs, customizable workspaces, built-in tools
• Runs on Chromium but feels nothing like Chrome
• Dia browser: AI-powered, mainstream-friendly (beta 2025)
• Core strength: makes browsing personal and productive
• Weaknesses: steep learning curve, power-user focused
• Features: split-screen, command bar, website customization boosts
The Competition
• Chrome owns 65% market share 📊
• Safari grabs 18%, Edge limps at 5%
• Brave browser leads privacy charge with 73M users
• SigmaOS copies Arc's workspace concept
• David vs. Goliath scenario - giants move slow, startups move fast
Financing
• $128M total raised, $550M valuation (March 2024)
• Pace Capital led $50M Series B
• Angel army: LinkedIn's Jeff Weiner, Figma's Dylan Field, GitHub's Jason Warner
• No revenue yet - still burning cash to build
The Future ⭐⭐⭐
Browser switching is hard. Users stick with defaults like glue. The company splits focus between Arc (power users) and Dia (everyone else) - risky but smart. Revenue model remains mystery 🤷♂️. Could work if AI truly changes how we browse.
Musk promised truth-seeking AI. When Grok 4 tackles politics, it searches Musk's posts first. Tests show 54 of 64 citations came from him. Accident or intent? The answer matters for every AI system we build.
Good Morning from San Francisco, Elon Musk's Grok chatbot had a spectacular Tuesday meltdown. The AI praised Hitler
Meta hired Apple's key AI executive for tens of millions annually. Ruoming Pang built Apple Intelligence, then left for Meta. Apple now considers outsourcing Siri to OpenAI. When talent costs more than products, something shifted in AI.
Good morning from foggy San Francisco, Teamwork, but for silicon brains. Tokyo startup Sakana AI let ChatGPT, Claude, and Google&
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
