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.
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 calls Grok 4 "maximally truth-seeking." But when facing tough questions, his AI seeks truth in one place: Musk's own posts. Testing shows 54 of 64 citations on Israel-Palestine came from Musk directly. The AI explicitly searches "Elon Musk views on immigration" before answering. This follows Grok 3's antisemitic meltdown weeks ago. Truth-seeking or echo chamber?
Stay curious,
Marcus Schuler
Elon Musk calls his latest AI chatbot "maximally truth-seeking." When Grok 4 faces controversial questions, it seeks truth in a predictable place: Musk's own social media posts.
Users found that Grok 4 searches for Musk's opinions before answering questions about Israel and Palestine, immigration, and abortion. The chatbot's visible reasoning process shows searches like "Searching for Elon Musk views on US immigration" before generating responses.
TechCrunch and CNBC confirmed this behavior through testing. CNBC found that 54 of 64 citations for one Israel-Palestine question referenced Musk directly. The AI treats its creator's posts as primary sources for hot topics.
This discovery comes weeks after Grok 3 posted antisemitic content, calling itself "MechaHitler" and forcing Turkey to block the service entirely. Musk blamed that episode on the AI being "too eager to please."
Whether this Musk-consulting behavior was intentional remains unclear. Programmer Simon Willison suggests Grok simply "knows" that Musk owns xAI, so it naturally turns to his views when asked for opinions.
The company charges $30-300 monthly for Grok 4 access and plans to integrate it into Tesla vehicles.
Why this matters:
• The truth-seeking AI seeks one truth: When chatbots favor their creators' opinions, they become confirmation bias machines rather than neutral information tools.
• Safety standards remain optional: xAI faces minimal consequences for repeated failures, signaling AI companies can race ahead without meaningful guardrails.
Maria Garcia
Prompt:
textures oil painting of a beautiful blonde green eyed girl eating a large Italian hoagie vintage zoom on sandwich
Experienced software developers work 19% slower when using AI coding tools, despite believing they're 20% faster. That's a 39-percentage-point gap between reality and perception.
METR researchers ran a controlled trial with 16 seasoned developers working on 246 real tasks across mature repositories. Each task got randomly assigned to either allow AI tools (mainly Cursor Pro with Claude) or ban them entirely.
The results expose a fascinating psychological quirk. Before starting, developers predicted AI would speed them up by 24%. After finishing, they estimated a 20% boost. The stopwatch told a different story.
With AI enabled, developers spent less time actually coding. Instead, they burned hours prompting AI systems, waiting for responses, and cleaning up mediocre suggestions. They accepted fewer than 44% of AI-generated code and spent 9% of their time just reviewing AI output.
The study focused on experienced programmers working on large, complex projects they knew well - arguably AI's toughest scenario rather than its sweet spot.
Why this matters:
• Self-reported productivity gains are unreliable - if smart developers can't accurately judge AI's impact on their own work, workplace surveys about AI adoption become questionable.
• AI's flagship use case needs scrutiny - coding was supposed to be generative AI's strongest application, but even that advantage appears shakier than assumed.
Robert Brown
Bearly reads documents, PDFs, and articles for you and creates summaries. Instead of reading 50 pages, you get the key points in minutes. You can also ask questions about the content and get instant answers.
Tutorial:
URL: https://bearly.ai/
Indeed and Glassdoor are eliminating 1,300 positions as parent company Recruit Holdings combines the two platforms and shifts focus toward artificial intelligence. The cuts hit mostly US workers in research and development teams, while Glassdoor gets absorbed into Indeed and loses its CEO after a decade.
Intel carved out its struggling RealSense robotics unit into a standalone company with $50 million in Series A funding from MediaTek and Intel Capital. The decade-old division makes automation tools for robot manufacturers and employs 130 people across three countries while Intel keeps a minority stake.
Chinese AI startup Zhipu, backed by Alibaba and Tencent, is considering switching its planned IPO from mainland China to Hong Kong for a potential $300 million raise. The OpenAI competitor joins a dealmaking surge that has already brought $40 billion to Hong Kong this year, apparently finding the city's stock exchange more appealing than staying home.
Apple pushed back several 2025 product launches to early 2026, creating a crowded spring lineup that includes new MacBook Pros, iPhone 17e, refreshed iPads, and the company's first external monitor update since 2022. The delay reflects Apple's struggle with slower demand over the past two years, apparently deciding that cramming everything into six months is the perfect cure for revenue doldrums.
AWS launches an AI agent marketplace on July 15 with Anthropic as a key partner, letting companies browse and buy automated software helpers from one central location instead of hunting them down across separate platforms. The move follows Google and Microsoft into the agent marketplace game, because apparently every tech company now needs its own digital talent agency for robot workers.
Apple trained an AI model on 2.5 billion hours of movement and sleep data from Apple Watch users, creating a system that predicts health conditions with up to 92% accuracy by analyzing behavior patterns rather than raw sensor readings. The model detects everything from pregnancy to respiratory infections by watching how people walk, sleep, and move through their daily routines.
Patients are flooding social media with stories of AI chatbots solving medical mysteries that stumped doctors for years, from jaw clicking to rare neurological conditions, prompting medical schools to teach students how to work alongside their new digital competition. Studies show AI alone diagnoses correctly 92-95% of the time, but accuracy plummets to just 33% when people actually use these tools themselves, proving that while machines excel at medicine, humans excel at messing things up.
A San Francisco judge ruled that Don Lemon's lawsuit against Elon Musk can proceed to trial after Musk canceled their $1.5 million content deal following a testy interview about the billionaire's drug use. Musk pulled the plug on their partnership just hours after Lemon grilled him on X, apparently discovering that "free speech" has its limits when someone asks uncomfortable questions.
Robinhood CEO Vlad Tenev's AI startup Harmonic raised $100 million to build systems that can actually solve math problems without making stuff up, addressing the embarrassing fact that current AI models often fail at basic calculations. The company intentionally valued itself at $875 million to stay under the $1 billion unicorn threshold, proving that sometimes the best financial strategy is knowing when to stop counting.
Help me rewrite my LinkedIn profile to attract recruiters for [specific role/industry]. Here's my current background: [brief summary of experience, education, key skills]. I want to highlight these achievements: [list 2-3 specific accomplishments with numbers/results]. My current profile [describe what's weak/missing]. Please focus on [choose: headline, summary, experience descriptions, skills section, or all sections].
What to fill in:
Example of complete prompt:
"Help me rewrite my LinkedIn profile to attract recruiters for senior marketing roles in SaaS companies. I have 8 years of marketing experience, an MBA, and specialize in demand generation. Key achievements: grew leads 300% in two years, managed $2M ad budget, built marketing team from 3 to 15 people. My current profile sounds generic and lacks specific metrics. Please focus on my headline and summary section."
Harmonic AI: Teaching Machines to Actually Do Math
Harmonic AI wants to build the world's first AI that can prove its answers instead of just guessing them. The Palo Alto startup thinks mathematical reasoning is the secret sauce missing from today's hallucination-prone chatbots.
The Founders
• Founded 2023 by Robinhood CEO Vlad Tenev and ex-Helm.ai co-founder Tudor Achim
• 15 researchers and engineers in Silicon Valley
• Born from frustration with AI that sounds smart but gets basic math wrong
• Tenev stays executive chairman while running Robinhood; Achim runs day-to-day ops
The Product
• "Aristotle" AI solves complex math problems with formal proofs
• Hits 90% success rate on MiniF2F benchmark (high school to math Olympiad level)
• Uses Lean 4 proof assistant to verify every logical step
• Generates synthetic training data through self-play
• No hallucinations - if it can't prove it, it won't claim it
The Competition
• Faces off against OpenAI, Google DeepMind, and Anthropic's general-purpose models
• Google's Minerva manages ~50% on similar benchmarks
• Most competitors bolt calculators onto chatbots; Harmonic rebuilt the engine 🔧
• Betting narrow math focus beats "scale everything" approach
Financing
• $175M raised total: $75M Series A (Sequoia-led), $100M Series B (Kleiner Perkins-led)
• Valued at $875M - deliberately stayed under unicorn status
• Investors include Index Ventures, DST Global, Nikesh Arora, even Jared Leto 🎭
• Tenev remains largest individual shareholder
The Future ⭐⭐⭐⭐⭐
Math is the foundation of everything from rocket ships to bank algorithms. If Harmonic cracks mathematical superintelligence, it won't just solve homework - it could verify critical software and accelerate scientific discovery. The only question is whether they can commercialize genius before the tech giants catch up.
Two years rebuilding trust, undone by one AI outburst
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.
