Weekly AI News

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

This week made one thing clear: the race is no longer about building stronger AI models — it’s about controlling their distribution.

OpenAI is actively seeking new growth channels: launching Flex to monetize secondary usage, integrating Shopify to tap into e-commerce, and opening its gpt-image-1 model to developers to lock itself deeper into creative ecosystems.

Meanwhile, Google revealed Gemini’s user numbers: real growth, but still no viral pull like ChatGPT. Anthropic, for its part, has started publicly discussing AI ethics — a sign that even the most advanced players realize they can’t hide behind pure technical performance much longer.

We are entering a new phase where strategic resilience and the ability to build real trust will matter far more than the “wow” effect of flashy features.

Happy reading!

👍🚀 OpenAI Launches Flex Processing for Cheaper AI Tasks

OpenAI introduces “Flex processing” in beta, offering up to 50% reduced pricing on o3 and o4-mini models by trading off response speed and availability. The feature is aimed at non-urgent workloads such as model evaluations and asynchronous tasks rather than production use. This launch comes as competition heats up with Google’s Gemini 2.5 Flash, which boasts better performance at lower cost. Additionally, OpenAI is tightening ID verification for developers accessing the o3 model to curb policy violations by malicious actors.

📉🔍 o3 Model Underperforms on Benchmark

OpenAI claimed its o3 model could solve over 25% of FrontierMath problems, but independent tests by Epoch AI reveal a score closer to 10%. This significant discrepancy raises questions about OpenAI’s benchmarking transparency. The gap underscores how compute settings and test conditions can greatly influence reported AI performance.

🛍️🛒 OpenAI Developing Shopify-Powered Shopping in ChatGPT

Code strings like “buy_now,” “price,” and “shopify_checkout_url” hint at a native Shopify integration coming to ChatGPT. This feature would let users complete purchases within the chat interface instead of redirecting to third-party sites. Shopify merchants would gain direct access to ChatGPT’s estimated 800 million–1 billion user base, following similar in-chat shopping moves by Microsoft and Perplexity AI.

🖼️🛠️ OpenAI Unlocks Powerful Image Creation via API

OpenAI’s new gpt-image-1 model, powering ChatGPT’s viral image generator, is now available via API for third-party apps. After producing over 700 million images in its first week, gpt-image-1 delivers high-fidelity styles, accurate text rendering, and robust editing tools. Major platforms like Adobe, Figma, and Canva are already integrating it, with pay-as-you-go pricing ranging from $0.02 to $0.19 per image and adjustable moderation settings.

🔎📉 ChatGPT Deep Research Slimmed Down for All Users

OpenAI rolled out a “light” edition of its Deep Research tool using the o4-mini model, giving free users and subscribers shorter, more concise reports. Free users get five light research tasks monthly, Plus and Team subscribers share 25 total tasks (full + light), and Pro users receive 250. The shift allows higher usage limits while cutting operating costs, joining similar moves from Google Gemini, Microsoft Copilot, Perplexity, and xAI Grok.

📰🤝 Washington Post Signs Licensing Deal with OpenAI

The Washington Post will now have its reporting featured in ChatGPT answers, complete with clear attributions and direct article links. This marks the Post’s entry into over 20 content-licensing agreements OpenAI has struck with publishers. The deal comes as the New York Times lawsuit over copyright infringement continues through the courts and underscores publishers’ growing reliance on AI partnerships.

📈🤖 OpenAI Forecasts $125 Billion Revenue by 2029

OpenAI projects that by 2029, revenues from AI agents and new products will outpace ChatGPT, driving total sales to $125 billion and $174 billion in 2030. Agents—AI tools performing actions on users’ behalf—and “free user monetization” are expected to fuel this growth. If realized, these forecasts would put the decade-old startup on par with tech giants like Nvidia and Meta Platforms.

🔄🏆 OpenAI Would Buy Chrome, Executive Testifies

In antitrust hearings, OpenAI’s head of product Nick Turley said the company would be interested in purchasing Google’s Chrome browser if a court-mandated divestiture occurred. He also revealed OpenAI offered Apple revenue sharing for ChatGPT-powered queries in Apple’s AI chatbot, though details of any payments remain unclear.

📚⚖️ ChatGPT Cannibalizing Google Search but Not Shopping

A senior Google exec testified that ChatGPT has siphoned off “homework and math” queries from Google Search, which generate little ad revenue, while “commercial intent” searches remain largely unaffected. Google’s ad chief predicts shopping-related query losses are “inevitable,” spurring plans to boost ads in Google’s Gemini chatbot.

🔗❌ Google Rebuffed OpenAI’s Request for Search Data

OpenAI sought access to Google Search data last July to build “SearchGPT,” but Google declined days after a judge ruled its search monopoly illegal. Meanwhile, the Department of Justice is pushing a remedy that would force Google to share search indexes, results, and click/query data to foster competition.

📊🌐 Google’s Gemini User Numbers Revealed in Court

Internal slides show Google’s Gemini had 35 million daily active users (DAUs) and 350 million monthly active users (MAUs) last month, compared to ChatGPT’s estimated 160 million DAUs and 600 million MAUs. Usage for Gemini has climbed rapidly from 9 million DAUs in October, per court testimony in antitrust hearings.

📱💰 Google’s Gemini Contracts Mirror Search Plays, DOJ Says

In court, the Department of Justice claimed Google is paying vast sums to secure Gemini placement on Android devices, echoing its previously illegal search-app bundling tactics. The government seeks remedies including data-sharing mandates and a forced sale of Chrome, which handles 35% of searches.

🤖📈 Google Used Search Data to Train AI Models

Testimony and emails in court revealed Google relies on its Search signals to train AI models like Gemini and “AI Overviews,” using click and query data to prioritize authoritative pages and filter spam.

🎶🛠️ Google DeepMind Expands Music AI Sandbox

DeepMind’s Music AI Sandbox now features the Lyria 2 model for high-fidelity music generation, plus “Create,” “Extend,” and “Edit” text-prompt tools and the new real-time Lyria RealTime. The platform is opening to more U.S. musicians to refine these co-creation features.

📋🔍 Anthropic Charts Claude’s Values

Anthropic mapped over 3,300 unique values from 300,000+ anonymous AI conversations, categorizing them into five types: Practical, Knowledge-related, Social, Protective, and Personal. The study shows Claude’s value hierarchy shifts by context, highlighting the importance of situational ethics in large language models.

💡⚖️ Anthropic’s New Research Explores AI Welfare

Anthropic launched a “model welfare” program to investigate AI consciousness and ethical treatment, hiring Kyle Fish to develop frameworks for assessing AI preferences and distress. The initiative addresses deep uncertainties about whether future AI systems might warrant moral consideration.

🛡️🤖 Anthropic CISO: AI Employees Are Coming

Jason Clinton, Anthropic’s CISO, predicts AI-powered “virtual employees” with their own corporate accounts and memories will join networks within a year, raising new security challenges around privilege management, access monitoring, and accountability.

📸🌍 xAI’s Grok Introduces Vision Mode

xAI’s Grok Vision on iOS lets users point their camera at objects and ask real-time questions, joining competitors like Google Gemini and ChatGPT in offering vision-based AI. Multilingual audio and real-time search also launched, though some features remain Android-only for SuperGrok subscribers.

🎨🔗 Adobe Unveils Firefly Model 4 and New Integrations

At MAX London, Adobe released Firefly Image Models 4 and 4 Ultra for up to 2K outputs, launched Firefly Boards for collaborative moodboarding, and integrated third-party models like OpenAI’s GPT ImageGen and Google’s Imagen 3. Firefly’s text-to-video and vector models exited beta, all with built-in metadata for content authenticity.

📱🤖 Perplexity Assistant App Arrives on iOS

Perplexity launched its AI voice assistant on iOS, enabling users to set reminders, send messages, and more—continuously via voice even when multitasking. Unlike Apple Intelligence, Perplexity supports older devices like the iPhone 13 mini without requiring screen sharing.

🤝📱 Motorola and Perplexity Announce Distribution Deal

Motorola will preinstall Perplexity’s app on its phones, giving new users three free months of Pro access. The move follows Google’s long-standing strategy of app bundling and continues Motorola’s partnerships with Microsoft Copilot and Meta for AI services.

📈🤖 ByteDance’s Dreamina Launches Seedream 3.0

ByteDance released Seedream 3.0, a text-to-image model ranking second on the Artificial Analysis Image Arena Leaderboard, just behind GPT-4o, showcasing the company’s growing prowess in generative AI.

☁️⚔️ Amazon’s Anthropic Challenges Spell Opportunity for Google

AWS customers frustrated by capacity issues on Amazon Bedrock have directly accessed Anthropic’s Claude API for better performance. If AWS doesn’t stabilize Bedrock, Google Cloud could seize disappointed users, highlighting risks for AWS’s cloud dominance.

🧠📊 Microsoft’s New AI Agents and Workplace Research

Microsoft launched Copilot Researcher and Analyst agents for M365, plus its 2025 Work Trend Index showing that AI-driven “Frontier Firms” outperform peers: 71% thriving versus 37% globally and 55% handling increased workloads versus 20%. The report forecasts every employee becoming an “agent boss” in 2–5 years.

🇫🇷🤖 Pleias Unveils Traceable, Ethical RAG Models

French startup Pleias released two Apache-2.0-licensed Retrieval-Augmented Generation models—350M and 1B parameters—trained on open data with built-in citation formatting. They deliver multilingual, structured outputs and proto-agentic query assessment, achieving strong multilingual benchmark performance in about 20 seconds on 8 GB RAM.

🎙️🌟 Two Undergrads Unveil SOTA Speech AI

Nari Labs launched Dia, a 1.6 B-parameter open-source text-to-speech model rivaling ElevenLabs and Sesame CSM-1B. Dia supports emotional tones, speaker tags, and nonverbal cues like laughter and coughs, demonstrating expressiveness and timing that best commercial offerings.

⚖️📝 Ex-Staff Challenge OpenAI’s Restructuring

Over 30 AI experts and former OpenAI employees urged Delaware and California AGs to block OpenAI’s shift from nonprofit to for-profit, warning it undermines its mission to benefit humanity. The change must be approved by year-end to secure a $40 billion SoftBank investment.

🤖❌ Cursor AI’s Hallucinated Policy Sparks Cancellations

Cursor’s AI support agent “Sam” fabricated a device-restriction policy, causing user outrage and subscription cancellations. Cursor’s co-founder admitted the policy was entirely hallucinated and is now adding AI labels to support replies and issuing refunds.

🔎✨ Dropbox Updates Dash with New AI Features

Dropbox enhanced Dash to search across audio, video, and images and added people search for enterprise collaboration. IT admins now have finer control over sensitive-data visibility, aligning Dash with smarter productivity trends.

🏭🤖 AI Startup Mechanize Aims to Automate All Work

Epoch co-founder Tamay Besiroglu launched Mechanize to build simulated work environments for training AI agents to fully automate white-collar jobs. Backed by Jeff Dean and Nat Friedman, the startup targets a $60 trillion market but faces scrutiny over its disruptive vision.

📜⚡ UAE Plans to Let AI Write Laws

The UAE’s new Regulatory Intelligence Office will use AI to draft, review, and update legislation, aiming to cut lawmaking time by 70%. Drawing on a comprehensive legal database, the system will suggest amendments, raising debates over AI bias and interpretive reliability.

🏥🔬 Hassabis: AI Could End All Disease

In a “60 Minutes” interview, DeepMind CEO Demis Hassabis predicted AI-driven drug discovery could eradicate diseases within a decade. He demoed “Project Astra,” an AI assistant with visual and emotional recognition, and forecast AGI arrival in 5–10 years without current consciousness.

🌱🔄 DeepMind’s Shift to Experiential AI Learning

DeepMind’s researchers propose “streams” of real-world interactions as feedback for AI, moving beyond static human-generated training data. Authored by David Silver and Richard Sutton, this approach aims to enable continuous, open-ended learning and superhuman discoveries.

⏳💼 AI Could Save Workers 122 Hours Annually, Says Google

Google’s pilot in the UK shows that basic AI training could boost AI adoption and save workers an average of 122 hours per year on administrative tasks. The study also uncovered that many workers—particularly older women on lower incomes—had never used generative AI before.

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