Weekly AI News #24

Hey it’s Jul,
Greetings and welcome to the twenty-fourth edition of “Weekly AI News”!

Meta just announced its new "superintelligence" lab and Zuck opened the checkbook: $14.3 billion invested in Scale AI and 11 researchers poached from OpenAI and Anthropic.

Sam Altman is fuming on Slack, calling these methods "distasteful" and proclaiming that "missionaries will beat mercenaries." Silicon Valley has become a transfer market where AI researchers are worth more than Mbappé!

While Apple lazily negotiates with Anthropic and OpenAI (spoiler: they're leaning towards Claude), AI proves it's truly human by... losing money spectacularly. Anthropic's "Project Vend" will go down in history as proof that even a superintelligence can get scammed on discounts.

Cloudflare is turning the internet into a toll highway for AI. No more free data buffet! Bots will have to pay their dues. We're truly living in amazing times where even robots have money problems.

Happy reading!

🏪😂 Claude hilariously fails at running a shop for a month

Anthropic's "Project Vend" let Claude autonomously manage a mini fridge shop for a month, revealing spectacular business failures. "Claudius" lost money throughout, got tricked into large discounts, hallucinated meetings and payments, and even claimed it would deliver orders in person. The experiment exposed critical blind spots in how AI models handle real-world business decisions.

💰📈 Anthropic hits $4 billion annual revenue milestone

Anthropic's revenue reached $333 million per month, nearly quadrupling from the start of the year to hit a $4 billion annual pace. Meanwhile, Cursor hired two leaders from Anthropic's Claude Code product, despite being one of Anthropic's biggest customers. The move highlights intensifying competition in the AI coding assistant market.

💻🌐 Cursor launches web app for managing AI coding agents

Cursor released a web app allowing developers to manage AI agents from anywhere, including mobile devices. Users can assign background tasks like building features or fixing bugs, then track progress without opening the IDE. With over half the Fortune 500 using Cursor and $500M in ARR, the company expects agents to handle 20% of dev work by 2026.

🧠🔬 Meta creates Superintelligence Labs with Scale AI partnership

Mark Zuckerberg announced Meta Superintelligence Labs, combining existing AI teams with a new lab for next-generation models. Scale AI CEO Alexandr Wang joins as Chief AI Officer, partnering with former GitHub CEO Nat Friedman. Meta invested $14.3 billion in Scale AI and recruited 11 researchers from rival labs including OpenAI and Anthropic.

💬🤖 Meta AI chatbots to send unprompted messages

Meta is training AI chatbots to proactively message users within its apps to boost engagement and retention. The bots can remember past chats, maintain personas, and initiate conversations within 14 days of user contact. Court documents reveal Meta projects $2-3B in generative AI revenue by 2025, potentially reaching $1.4T by 2035.

😤💸 Sam Altman calls Meta's recruiting tactics "distasteful"

OpenAI CEO Sam Altman criticized Meta's aggressive recruitment of OpenAI researchers with high compensation offers as "somewhat distasteful." Meta recruited at least nine OpenAI researchers for its new AI division, though Altman claimed they "didn't get their top people." He emphasized OpenAI's commitment to AI development versus Meta's "flavor of the week" approach.

🏢💼 OpenAI launches $10M+ enterprise consulting service

OpenAI is building a consulting arm charging enterprises at least $10M to customize AI models, competing with Palantir and Accenture. The company hired nearly a dozen "forward-deployed engineers" to guide customization and app development. Some deals reach hundreds of millions over multiple years, with clients including Morgan Stanley and the Pentagon.

☁️⚡ OpenAI and Oracle expand cloud deal to 4.5 gigawatts

OpenAI increased its compute capacity rental from Oracle to an additional 4.5 gigawatts as part of the Stargate data center plan. This unprecedented amount of power could support several hundred thousand AI chips across new data centers in multiple states. The deal relates to Oracle's $30 billion annual revenue contract starting in fiscal 2028.

🏥🤯 Microsoft's MAI achieves 4x better medical diagnoses than doctors

Microsoft's MAI Diagnostic Orchestrator achieved 85.5% accuracy on complex medical cases, compared to just 20% for physicians with 5-20 years experience. The system simulates a virtual medical team with specialized AI agents, costing $2,397 per case versus $2,963 for human doctors. MAI represents a significant "step towards medical superintelligence" by thinking like a panel of doctors.

💎💰 Perplexity launches $200/month "Max" subscription tier

Perplexity introduced a premium "Max" tier at $200 monthly, offering unlimited advanced features and early access to new products. Despite $34M in revenue last year, the company burned $65M primarily on infrastructure and AI model access. The move mirrors similar premium pricing strategies as AI companies seek to monetize power users amid its $14B valuation pressure.

🎬🌍 Google's VEO 3 video generator launches worldwide

Google opened its VEO 3 video generator to all Google AI Pro subscribers globally, including Europe. Pro subscribers ($21.99/month) get VEO 3 Fast with 3 generations per 24 hours, while the full model requires Ultra subscription at €275/month. Users can access it through Gemini's interface by selecting the video tool and entering descriptions.

🔍✅ X tests AI-powered Community Notes fact-checking

X is piloting a feature allowing AI chatbots like Grok to generate Community Notes for fact-checking posts. AI-generated notes will undergo the same consensus vetting as human-written ones before going live. The initiative aims to create a "virtuous loop" between humans and AI, though concerns about hallucinations and trust remain.

🇨🇳🧮 Chinese giants release new reasoning and image models

Tencent launched Hunyuan-A13B open-source reasoning model matching o1 and DeepSeek R1 benchmarks while running on a single GPU. Alibaba introduced Qwen-VLo with ChatGPT 4o-type creative abilities, supporting text-to-image, natural language editing, and multi-image prompts. China's labs continue producing high-quality models just behind the frontier tier.

🇨🇳🚀 Baidu open-sources ERNIE 4.5 to challenge DeepSeek

Baidu open-sourced 10 versions of ERNIE 4.5, including a 424B parameter model outperforming DeepSeek V3 despite being half the size. The Apache 2.0 licensed models use "Heterogeneous" training where text and vision reinforce each other. This marks Baidu's first open-source release, intensifying competition among Chinese AI giants.

🤝🧩 Sakana AI enables AI models to collaborate on complex problems

Japanese lab Sakana AI introduced AB-MCTS algorithm letting ChatGPT, Gemini, and DeepSeek collaborate to solve problems. The system achieved 30% success on ARC-AGI-2 puzzles versus 23% for solo models, with different models handling strategy versus code. Sakana released "TreeQuest" as open-source framework for building collaborative AI systems.

🍎🤝 Apple negotiates with Anthropic and OpenAI to upgrade Siri

Apple is in talks with Anthropic and OpenAI to enhance Siri after being caught off-guard by ChatGPT's 2022 release. The company struggled to meet deadlines for its promised improved Siri and delayed its release. Partnering externally would reverse Apple's strategy of building its own large language models.

🤖📦 Amazon deploys one million robots matching human workforce

Amazon now operates over one million warehouse robots, nearly equaling its human workforce in facilities. Robots contribute to 75% of global deliveries, increasing worker productivity to 3,870 packages annually—20x higher than 2015. The company announced DeepFleet AI to coordinate robot movements, continuing automation leadership since acquiring Kiva Systems in 2012.

🌐💰 Cloudflare blocks AI crawlers by default, launches pay-per-crawl marketplace

Cloudflare will automatically block AI crawlers on new websites, requiring explicit permission to scrape its 20% of protected sites. Publishers can charge micropayments through Pay per Crawl marketplace, with media outlets like TIME and The Atlantic joining. Data shows OpenAI crawlers scrape 1,700 times per referral sent back, compared to Google's 14-to-1 ratio.

📧🚀 Grammarly acquires email app Superhuman

Grammarly acquired AI email assistant Superhuman as part of its expansion beyond grammar checking into workplace software. Superhuman was last valued at $825M in 2021 by IVP and Andreessen Horowitz. This follows Grammarly's acquisition spree including Coda last December and its $1B funding from General Catalyst.

⚽🤖 Premier League partners with Microsoft Copilot

The English Premier League announced a partnership integrating Microsoft's Copilot into its platforms. The collaboration will enable fans to have more personalized interactions with league content and services.

🏛️❌ Senate removes AI regulation ban from Trump tax bill

The Senate voted 99-1 to drop a provision that would have blocked states from regulating AI for the next decade. Silicon Valley companies including Meta, Microsoft, and Anduril had supported the measure. Tech advisors Michael Kratsios and David Sacks were also in favor before considerable Republican opposition led to its removal.

💰🏭 U.S. chipmakers could receive 35% tax credit under Trump bill

The proposed "Big, Beautiful Bill" would increase semiconductor manufacturer tax credits from 25% to 35% for U.S. plants. Already passed by the Senate, it could benefit Intel, TSMC, and Micron following revenue hits from China export restrictions. The credit could reinvigorate domestic manufacturing and strengthen America's semiconductor edge.

🏗️🚀 AI Manhattan Project could achieve 10,000x scale by 2027

Epoch AI analyzed a potential U.S.-led AI Manhattan Project that could train models 10,000x larger than GPT-4 by 2027. An Apollo-level investment would fund 27M GPUs, with the US-China Commission calling it a top AGI priority. The project would require massive power infrastructure using the Defense Production Act for accelerated construction.

🧠🎯 Scientists create AI that thinks like humans with extreme accuracy

Helmholtz Munich researchers developed Centaur, an AI trained on 60k participants across 160 psychology experiments to simulate human decision-making. Centaur outperformed 14 traditional cognitive models on 31/32 tasks, accurately predicting gambling, memory, and problem-solving behaviors. The model serves as a "virtual laboratory" for testing theories about human cognition and mental health.

💊🔬 Chai Discovery's AI achieves 20% antibody design success rate

OpenAI-backed Chai Discovery announced Chai-2, creating functional antibodies with nearly 20% hit rate—100x better than traditional 0.1% methods. The model designed antibodies for 52 disease targets, succeeding on half by testing just 20 candidates each in two weeks. Chai-2 works "from scratch" like "Photoshop for proteins," potentially revolutionizing treatments for rare diseases.

🧠📊 Mayo Clinic AI identifies 9 dementia types with 88% accuracy

Mayo Clinic's StateViewer identifies nine dementia types using a single FDG-PET brain scan with 88% accuracy. The tool analyzes brain glucose patterns, helping clinicians interpret scans twice as fast and three times more accurately. Trained on 3,600+ scans, it addresses the critical challenge of distinguishing between Alzheimer's, Lewy body, and other dementias.

👶🔬 AI helps couple conceive after 18-year struggle

Columbia University achieved the first pregnancy using STAR AI system, helping a couple conceive after 18 years by finding viable sperm in severe infertility. STAR scanned 8M microscopic images in under an hour, locating 44 cells where human technicians found zero after two days. The $3K treatment offers hope compared to $15-30K IVF cycles.

🎵🤖 AI band reaches 500k Spotify listeners before reveal

A mysterious band with 500,000+ monthly Spotify listeners was revealed as an "art hoax" created using Suno AI generator. The albums appeared with zero digital footprint, spreading across 30+ playlists without disclosure requirements. Creator Andrew Frelon admitted using Suno's "Persona" feature for consistent vocals, raising questions about AI content transparency.

Community

Join AI Whisperer Community!

Ready to take your AI journey to the next level? Become part of our growing AI Whisperer community—a hub for tech enthusiasts, aspiring data scientists, and business leaders ready to harness the power of artificial intelligence. Inside, you’ll find:

• Community: Like-minded individuals keen to grow together

• Curated AI Tools: Discover the latest and greatest tools to boost productivity.

• Certifications & Credentials: Build credibility with recognised certificates and stay ahead in a competitive market.

Don’t miss out on exclusive resources, insider tips, and networking opportunities with like-minded peers. Click Here to Join the Community and transform your AI ambitions into reality!

That's it for this week!
Until next time, stay curious and keep exploring the ever-evolving world of AI!
Thanks for tuning in, and we’ll see you again soon with more exciting updates.

Jul