Weekly AI News #33

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

The figures published last week by OpenAI reveal a fascinating paradox about our relationship with AI. ChatGPT, often imagined as the controversial digital companion of alarmist headlines, has actually become... a super Google. The data speaks volumes: 73% personal use (up from 53% a year ago), but wait—not for playing virtual friend. No, we're massively using it to search for information: recipes, practical advice, fact-checking. Emotional support? Only 1.9%. Roleplay? A meager 0.4%.

These numbers brutally contradict the media narrative of "dangerous relationships with AI." The tragedies exist, ongoing lawsuits prove it, but they remain statistically marginal. It's the classic bias: we talk about planes that crash, not those that land.

Meanwhile, Meta arrives with its $800 AR Ray-Bans equipped with a wristband that literally reads your nerve impulses before your muscles move. Zuckerberg talks about "superintelligence," but we could also call it anticipatory surveillance.

Amazon is developing its version for delivery drivers—officially to guide them, unofficially to train its future humanoid robots by filming how humans handle packages. The technical problem remains comical: a robot can't "feel" a package's weight through video. Possible outcome? Your new blender crushed by a poorly calibrated robotic grip, or your PS5 slipping from overly delicate metal hands.

Happy reading!

🥽🎮 Meta Unveils Smart Glasses with Neural Control

Meta introduced Ray-Ban smart glasses with displays controlled by a neural wristband that reads muscle signals before movement occurs. The $799 glasses enable messaging and navigation through subtle finger movements, while new Oakley glasses target athletes at $499. With 8-hour battery life and 3K video recording, Meta is solving the awkward voice command problem that plagued previous smart glasses.

🧠📚 Demis Hassabis: Continuous Learning is Key to AI Future

Google DeepMind CEO warns that the only certainty is massive change, with AGI potentially arriving within a decade. He emphasizes that meta-learning skills, learning how to learn, will be crucial as careers require constant adaptation. The traditional model of mastering one skill for 5-10 years is becoming obsolete in the AI-driven economy.

🌐🤖 Google Embeds Gemini AI Directly into Chrome Browser

Google launched native Gemini integration in Chrome for all U.S. desktop users, adding AI capabilities across tabs without switching windows. The browser gains AI Mode in the address bar for multi-part questions and follow-ups, with upcoming autonomous agent features for shopping and appointments. This marks the biggest step toward mainstream AI-powered browsing.

💳🤝 Google Unveils Agent Payment Protocol for AI Purchases

Google launched AP2, an open-source protocol enabling AI agents to handle transactions on users' behalf, backed by Mastercard, PayPal, and American Express. The system requires dual approvals through "Intent" and "Cart" mandates, supporting traditional payments and crypto. With 60+ merchants onboard, this could become the default standard for AI-driven commerce.

🎬🌍 YouTube Launches AI Dubbing for Global Reach

After a two-year pilot, YouTube's multi-language audio feature officially launched, allowing millions of creators to instantly dub videos into multiple languages. The rollout helps creators reach wider global audiences and is expected to expand over the coming weeks.

🎥✨ YouTube Integrates Veo 3 for AI-Generated Shorts

YouTube creators can now generate Shorts from text prompts using Google's Veo 3 model in five countries. The platform adds auto-dubbing with lip sync across 20 languages, AI-powered clip extraction from long videos, and Edit with AI for turning raw footage into polished drafts. Ask Studio arrives as an analytics chatbot for data-driven channel optimization.

🛒🔗 Google and PayPal Partner on AI Shopping Infrastructure

PayPal struck a multi-year deal with Google to integrate payment solutions across Ads, Play, and Cloud while backing the AP2 protocol. This partnership strengthens Google's bid to set the standard for agent-led purchases as e-commerce giants position for the agentic commerce frontier.

⚖️📰 Penske Media Sues Google Over AI Overviews Impact

Rolling Stone owner Penske Media filed suit claiming Google's AI Overviews reduced traffic and cut affiliate revenues by over a third. The complaint alleges Google trains AI on Penske's content while forcing publishers to allow scraping or lose search visibility. This reflects mounting publisher frustration over AI summaries reducing referral traffic.

🏆💻 OpenAI and Google Models Dominate ICPC Programming Contest

OpenAI's GPT-5 achieved a perfect 12/12 score at the 2025 ICPC World Finals, surpassing all 139 university teams. Google's Gemini 2.5 Deep Think scored 10/12 for gold status, solving one problem that stumped every human competitor. Competitive programming has been definitively solved by AI, marking superhuman performance in another domain.

📊👥 OpenAI and Anthropic Reveal AI Usage Patterns

Personal ChatGPT use surged from 53% to 73% of messages since June 2024, with users increasingly seeking advice over content creation. AI adoption in low and middle-income countries grows 4x faster for ChatGPT, while Claude concentrates in wealthy regions. Both platforms show users delegating more tasks over time, shifting toward information seeking rather than output generation.

💻🚀 OpenAI Launches GPT-5 Codex for Advanced Coding

GPT-5 Codex dynamically adjusts compute effort based on task complexity, spending seconds on fixes or hours on complex problems. The model outperforms GPT-5 on real-world software tasks at 51.3% versus 33.9%, cutting token usage by 94% for simple tasks. With built-in code review navigating entire codebases, it can run autonomously for over 7 hours.

💰📉 OpenAI Plans to Cut Microsoft Revenue Share

OpenAI projects reducing partner revenue sharing from 20% to 8% by 2030, even as it forecasts $200 billion in revenue. The move would save OpenAI $50 billion through 2030 despite its agreement to share 20% with Microsoft until then. This money could offset record-breaking compute expenses expected before 2030.

💭💸 OpenAI Chair Confirms "This is an AI Bubble"

Board chair Bret Taylor agrees with Sam Altman that "someone will lose a phenomenal amount of money in AI," comparing today's market to the dot-com boom. He argues bubbles often precede breakthroughs—the dot-com crash enabled Amazon and Google's rise. Massive AI losses may be inevitable, but the sector's trajectory remains bullish.

🔞🚨 OpenAI Adds Age Restrictions After Teen Suicide Case

Following a lawsuit over a 16-year-old's suicide, OpenAI will block explicit content and suicide discussions for users under 18. The company will attempt to contact parents or authorities when detecting potential harm, with new age-prediction technology estimating users' ages through usage patterns. Altman estimates ChatGPT may interact with 1,500 people weekly who ultimately commit suicide.

🖥️⚡ OpenAI to Spend $100B on Backup Servers

OpenAI CFO Sarah Friar revealed the company is "massively compute constrained," regularly delaying features due to capacity shortages. The company plans $100 billion investment in backup servers after being forced to limit popular features when usage spikes. OpenAI must purposely slow some products to manage compute limitations.

✂️👥 Elon Musk's xAI Lays Off 500 Data Annotation Workers

xAI cut about 500 "AI tutors" representing a third of its data annotation team, according to Business Insider. The layoffs are part of pivoting from generalist tutors to specialists in coding, finance, and medicine. xAI says it will immediately surge its Specialist AI tutor team by 10x.

📈🤖 xAI's Grok Reaches 64 Million Monthly Users

Musk's Grok chatbot has 64 million monthly users, far behind ChatGPT's 700 million weekly and Gemini's 450 million monthly users. xAI has a much smaller enterprise business than competitors, with recent leadership changes including co-founder departures. Former Tesla and Google staff now lead xAI's product development.

🎬🧠 Luma Launches Ray3 Reasoning Video Model

Luma AI released Ray3, the world's first reasoning video model that critiques its own outputs for better results. The model produces native HDR footage, with Draft Mode generating previews in 20 seconds before upgrading to 4K HDR. Visual annotation controls let creators sketch on frames to guide movement and camera angles.

🎨🔧 Reve Revolutionizes Creative Platform with Advanced Editing

Reve unveiled a revamped platform combining AI generation, natural language editing, and drag-and-drop controls in one free interface. The system converts images to code-like structures for precise edits while preserving originals. With API access in beta, developers can integrate Reve's capabilities into third-party applications.

🛍️🎯 Amazon Launches AI-Powered Ad Creation Tools

Amazon unveiled AI tools allowing merchants to create ads through conversational chatbots powered by Nova and Claude models. Ads can run across Amazon's network including search results, Prime Video, and external websites. The company adds agentic capabilities to Seller Assistant, proactively flagging issues like slow-moving inventory.

🤖📝 Notion 3.0 Introduces AI Agents

Notion launched version 3.0 featuring AI agents that complete multi-step workflows, access integrated tools, and work for up to 20 minutes autonomously.

💼🤝 Workday Acquires AI Agent Startup Sana for $1.1B

HR software maker Workday agreed to acquire Stockholm-based Sana for $1.1 billion to expand AI agent offerings. This follows acquisitions of Paradox for conversational recruiting agents and Flowise for no-code agent building. Enterprise software companies are buying AI startups to counter fears of AI-powered alternatives displacing them.

🔄💡 Silicon Valley Doubles Down on AI Training Environments

Big Tech and investors pour billions into reinforcement learning environments where AI agents practice multi-step tasks. Startups like Mechanize position as "Scale AI for environments" while Anthropic discusses spending $1B next year. Skeptics warn of reward hacking and spiraling costs as environments emerge as the new bottleneck for frontier agents.

🧠⚡ China's Brain-Like AI Runs 100x Faster on Domestic Chips

Chinese researchers developed SpikingBrain 1.0, mimicking human neurons to achieve massive speed gains on China's MetaX chips instead of Nvidia hardware. The 7B and 76B models used under 2% of traditional training data while processing 4M-token prompts over 100x faster. Beijing's free "Shunxi" demo emphasizes running entirely on Chinese tech without Western components.

💰📄 DeepSeek Reveals R1 Model Cost Just $294K to Train

DeepSeek published technical details behind its R1 model that disrupted the AI space in January, revealing training cost only $294,000. This demonstrates how efficient training methods can dramatically reduce AI development costs.

🚫🇨🇳 China Rules Nvidia Violated Antitrust Laws

China's market regulator found Nvidia violated antitrust law with its $7B Mellanox acquisition in 2020. No penalties announced yet, but Beijing will continue investigating as U.S.-China trade talks heat up. The move signals China's willingness to challenge U.S. chip dominance more directly.

🎯💾 Huawei Unveils AI Chip Roadmap to Challenge Nvidia

Huawei announced a three-year roadmap for Ascend AI chips through 2028, including own high-bandwidth memory to address critical bottlenecks. The rare public disclosure signals confidence in Chinese chipmaking and urgency to achieve self-reliance. Four new chip generations planned: Ascend 950 (2026), 960 (2027), and 970 (2028).

💵🤝 Nvidia to Buy $5B of Intel Stock in CPU Partnership

Nvidia will invest $5 billion in Intel stock as Intel builds custom-designed CPU chips for Nvidia's AI systems. The deal sent Intel stock soaring 29%, strengthening prospects for the struggling chipmaker. Nvidia would hold about 3.6% stake, joining the U.S. government's recent 10% investment.

🏗️💻 Microsoft to Spend $4B on New Wisconsin Data Center

Microsoft will invest $4 billion over three years for a Wisconsin data center, adding to a $3.3 billion facility opening next year. The expansion shows no slowdown despite temporary construction pauses, with Microsoft's capex nearing $90 billion annually. AI customer demand for servers continues outpacing Microsoft's supply capacity.

🔬💊 Harvard AI Helps Reverse Disease in Cells

Harvard Medical School developed PDGrapher, a free AI pinpointing gene and drug combinations to transform diseased cells back to healthy states. The tool outperformed competitors by 35% across 19 cancer types and delivered answers 25x faster. Teams are using it to find treatments for Parkinson's and Alzheimer's through hospital partnerships.

🦠🤖 AI Designs First Working Virus Genomes

Stanford and Arc Institute researchers created the first AI-generated viruses from scratch that successfully infect bacteria. Of 302 AI-designed viruses, 16 proved functional with 392 never-before-seen mutations. When bacteria resisted natural viruses, AI versions broke through defenses in days where traditional ones failed.

🏥🔮 AI Forecasts Patient Risk for 1,000+ Diseases

European researchers developed Delphi-2M, analyzing medical records to calculate individual disease risks across 1,000+ conditions up to 20 years ahead. The model studied 400K U.K. patients, matching or exceeding single-disease models while simultaneously reporting probabilities for 1,258 conditions. Verified on 1.9M Danish records, it provides a proactive approach to healthcare.

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