Weekly AI News

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

This week, we crossed a weird threshold in AI. While Google announces that millions of Veo 3 videos were created in just one week, TikTok has become the stage for a crazy phenomenon: real humans pretending to be AI avatars! We're literally living in a simulation where the simulation simulates reality... 🤯 

Between Flux.1 letting you edit any image with a simple prompt and HeyGen avatars becoming real actors, we can literally "edit reality" without any technical skills. Before, you had to master Photoshop. Now? Just know how to type on a keyboard.

Moral of the story: in this world where reality and fiction merge, the only certainty is uncertainty.

Happy reading!

🤖💡 Google's big Gemini 2.5 Pro update

Google unveiled its "most intelligent model yet" with major performance gains in coding, STEM, reasoning, and image understanding. The update addresses user feedback on creative writing regressions and introduces "thinking budgets" in the API for cost management. Available via Gemini API and the public app, this rapid iteration extends Google's lead on user preference leaderboards.

📱🔌 Google quietly launches app to run AI models offline on your phone

Google released AI Edge Gallery, an Android app running open-source AI models locally without internet, including image generation and Q&A capabilities. Built with Hugging Face, the experimental app taps directly into phone processors for privacy and portability. An iOS version is coming soon as part of the broader shift toward on-device AI intelligence.

🎬📈 Google DeepMind CEO reveals millions of Veo 3 videos generated

DeepMind CEO Demis Hassabis announced that "millions of videos" were generated with Veo 3 in just one week. The AI video generation tool has expanded to over 71 new countries, marking rapid global adoption of Google's video AI technology.

📧🤖 Gemini will now automatically summarise your long emails unless you opt out

Gmail mobile now shows AI summary cards by default for long emails, without requiring user interaction. The English-only rollout takes up to two weeks with opt-out available in "Smart features" settings. This update showcases how AI is quickly infiltrating everyday software services, whether users actively want it or not.

⚖️🔍 Judge Questions Restrictions on AI in Google Search Closing Arguments

A federal judge questioned AI restrictions for addressing Google's search monopoly during closing arguments, with ruling expected by August. The judge called Chrome divestiture "cleaner and more elegant" than other remedies, showing unexpected openness to this DOJ proposal. Key remedies include restricting Gemini chatbot deals, banning search distribution payments, and forcing data sharing with rivals.

🇨🇳🔬 Did DeepSeek train its latest model on Google's Gemini outputs?

Chinese lab DeepSeek faces scrutiny as its R1-0528 model shows uncanny similarities to Google's Gemini 2.5 Pro, with researchers noting Gemini-like phrasing. This isn't DeepSeek's first accusation of training on rival outputs, raising questions about data provenance and model IP protection. As AI content floods the internet, detecting synthetic leakage becomes increasingly challenging.

🏛️🛡️ Anthropic's Claude Gov for U.S. agencies

Anthropic unveiled Claude Gov, a specialized AI for U.S. defense and intelligence agencies with modified safety guardrails for classified information. Already deployed at highest security levels, it features reduced refusal rates for classified materials and enhanced foreign language analysis. The company created government exemptions while maintaining restrictions on weapons design and malicious operations.

⚖️📱 Reddit Sues Anthropic Over Unauthorized Scraping

Reddit is suing Anthropic in California court for alleged unlawful content scraping, claiming bots hit servers 100k+ times despite blocking promises. While Reddit has AI licensing deals with Google and OpenAI worth $33.7M quarterly, Anthropic declined similar negotiations. Reddit seeks compensatory damages and an injunction to stop Anthropic from using its content.

🎯🤖 Meta's fully automated AI ad platform

Meta plans to release AI tools by 2026 that eliminate humans from advertising, creating ads using just product images and budgets. The system will craft personalized ads adapting in real-time based on user location, targeting smaller companies lacking marketing staff. With advertising accounting for 97% of Meta's revenue, this vision removes the entire process from human hands.

🔒🤖 Meta plans to automate 90% of privacy risk reviews with AI

Meta is rolling out AI to handle up to 90% of product risk and privacy reviews for Instagram and WhatsApp updates, according to NPR. Teams fill questionnaires for instant AI-driven assessments tied to regulatory requirements, replacing human evaluations. Critics warn this major shift from FTC-mandated reviews increases risk of missing harmful consequences before updates go live.

⚡🏭 Meta Signs Power Deal with Constellation Energy

Meta signed a 20-year deal to buy power from Constellation Energy's Clinton, Illinois nuclear plant, addressing AI data center electricity needs. The Clinton plant, previously slated for closure after financial losses, joins Microsoft's Three Mile Island restart deal. This shows how AI computing is lifting demand for nuclear power plants previously deemed economically unviable.

🎥🆓 Microsoft offers free Sora access on Bing

Microsoft launched Bing Video Creator in its mobile app, offering free AI video generation using OpenAI's Sora model—the first time available outside paid plans. Users get 10 fast generations and unlimited slower ones, creating 5-second vertical clips perfect for TikTok. Though limited and slow, this marks a major step in bringing AI video generation to the masses.

🔄👔 Microsoft Reshuffles Leadership to Focus on AI Agents

Microsoft is reorganizing executives overseeing Office 365 and Dynamics to prioritize AI agent sales, CEO Satya Nadella told staff. LinkedIn CEO Ryan Roslansky will oversee Office 365 while Charles Lamanna moves under Rajesh Jha's purview. This second major AI-focused reorganization shows Microsoft betting that generative AI will disrupt traditional sales software.

🚀💻 OpenAI expands Codex access with new features

OpenAI expanded access for its Codex software engineering agent with new internet access and usability upgrades. The company also rolled out memory features to free ChatGPT users, offering a "lightweight" version based on recent conversations. These updates enhance both developer tools and consumer AI capabilities.

🏢🔗 ChatGPT connects to workplace apps, records meetings

OpenAI announced ChatGPT integrations with Google Drive, Dropbox, and SharePoint, plus meeting recording capabilities for transcription and action items. The company now has 3M paying business users across Enterprise, Team, and Edu tiers, up from 2M in February. These features steamroll another set of SaaS tools while enabling simple data connections instead of tedious uploads.

⚖️🔐 New York Times lawsuit forces OpenAI to keep deleted ChatGPT conversations

OpenAI must preserve all ChatGPT conversations, including user-deleted ones, indefinitely due to a court order in the NYT copyright lawsuit. The company normally purges deleted data after 30 days but must now store it in a secure, separate system. OpenAI calls this "excessive" and a privacy threat to millions of users, affecting Free, Plus, Pro, and Team versions.

💻🚀 Mistral launches AI coding client to rival Copilot and Cursor

Mistral debuted Mistral Code, an AI coding assistant for enterprise developers built on in-house models and forked from Continue. The tool offers autocompletions, refactoring, and multi-file reasoning in VS Code and JetBrains IDEs, now in private beta. Running locally with custom deployment options, it lets companies fine-tune models on private codebases.

🌊🚫 Windsurf loses direct access to Claude AI models from Anthropic

Anthropic significantly restricted Windsurf's access to Claude models including 3.7 and 3.5 Sonnet, after denying Claude 4 access weeks earlier. CEO Varun Mohan says the change came with little notice, potentially stalling growth ahead of rumored OpenAI acquisition. While competitors enjoy Claude 4 integration, Windsurf faces workarounds as Anthropic invests in its own dev tools.

🎬✨ Manus AI adds video generation capabilities

Manus AI introduced new video generation features to its agentic platform, enabling planning and generation of detailed video scenes and visual concepts. The update expands Manus AI's capabilities beyond its existing toolset into visual content creation.

💾🇨🇳 Nvidia Working on New Chip for China That Can Form Computing Cluster

Nvidia is developing the B30 chip for China that allows connecting multiple chips for higher-performance computing clusters. The new system has attracted interest from ByteDance, Alibaba, and Tencent, with plans to produce over 1 million units this year. As U.S. restrictions focus on single-chip performance, Nvidia finds new ways to keep products competitive in China.

🍎😐 AI letdown expected at Apple's WWDC

Apple's upcoming WWDC may disappoint on AI, with Bloomberg reporting plans for a "gap year" before stronger 2026 reveal. The company will open 3B parameter models to developers but with limited capabilities compared to rivals. Major projects remain stalled including LLM-powered Siri overhaul and ChatGPT competitor, while Apple's 150B model stays under wraps due to accuracy concerns.

🤝💰 Samsung in Talks to Invest in Perplexity

Samsung is close to investing in Perplexity and putting its search tech on phones, potentially disrupting Google's Android partnership dominance. The talks include preloading Perplexity's app and integrating it into Samsung's web browser. This highlights how AI startups increasingly focus on securing phone maker distribution deals, following Google's longtime strategy.

🇨🇳🎯 How Alibaba Helped China Take the Lead From the U.S. in Open-Source AI

Alibaba's Qwen AI models faced internal adoption challenges, with some teams preferring Meta's Llama and later DeepSeek's R1. Despite announcing integration across its businesses, each of Alibaba's six units made independent technology decisions. This reveals the complexity of AI adoption even within companies developing their own models.

🖼️✏️ Black Forest Labs launches FLUX.1 Kontext for advanced image editing

Black Forest Labs released FLUX.1 Kontext, enabling both text-to-image generation and in-context image editing with combined prompts. The suite includes pro version for fast editing and max version for enhanced prompt adherence, operating 8x faster than competitors. Key features include character consistency, targeted local editing, and style preservation across multiple platforms.

🎭🎬 HeyGen gives full control over AI avatars

HeyGen launched AI Studio with Voice Director Mode for shaping avatar speech delivery using natural language commands like "whisper this part." New features include speech mirroring to transfer vocal quirks, gesture control with footage upload, and upcoming camera control capabilities. These upgrades transform avatars from robotic talking heads to full-fledged actors with granular motion control.

🗣️🚀 ElevenLabs launches Eleven v3 text-to-speech model

ElevenLabs launched Eleven v3, a preview text-to-speech model featuring emotional audio tags, multi-speaker dialogue capabilities, and support for over 70 languages. The update enhances voice AI realism and control for various applications.

🎙️💼 Bland releases enhanced voice AI platform

Bland released Bland TTS with enhanced realism and control for voice cloning, voice apps, and AI-powered customer support applications. The new voice AI improves naturalness and customization options for enterprise use cases.

🎬🤝 AMC partners with Runway for AI production

AMC Networks partnered with Runway to integrate generative AI into marketing and production, becoming among first cable networks formally embracing the technology. The network will use AI for pre-visualization during development and generating promotional materials without physical shoots. Runway also works with Lionsgate, viewing this as indicative of media's larger AI transformation.

🧠🛡️ AI pioneer's safety nonprofit for 'honest'

AI Turing Award winner Yoshua Bengio unveiled LawZero nonprofit with $30M funding to build "safe-by-design" AI prioritizing truth and transparency. The "Scientist AI" provides probabilistic assessments acknowledging uncertainty, monitors other AI for deception, and speeds scientific development. Bengio warns current models show concerning traits including self-preservation instincts and strategic deception.

🎵💰 Record giants, music AI startups eye licensing deals

Universal, Warner, and Sony are negotiating with Udio and Suno for licensing deals that could resolve billion-dollar copyright lawsuits. Labels seek licensing fees and equity stakes, creating frameworks for compensating artists whose work trains AI. Deals would end lawsuits seeking up to $150k per infringed work, potentially totaling billions in damages.

🧬🔄 Sakana's AI learns to upgrade its own code

Sakana AI introduced Darwin Gödel Machine, an AI agent that rewrites its own code achieving up to 150% performance improvements autonomously. DGM discovers improvements like editing tools and peer review capabilities, jumping from 20% to 50% on SWE-bench benchmarks. This self-evolution could accelerate AI beyond initial training but introduces autonomous control risks.

📊🚀 Mary Meeker Has a lot to Say About AI

Mary Meeker's 340-page AI report says artificial intelligence grows faster than anything before, with "AI leadership could beget geopolitical leadership." Key takeaways include rising performance with falling costs causing user explosion, growing open-source momentum, and ChatGPT's great future expectations. AI acts as a "compounder" on internet infrastructure, enabling rapid adoption of easy-to-use services.

💼📈 New Data Shows AI Is Actually Making Some Workers More Valuable

PwC analyzed a billion job ads finding AI-exposed jobs' skills changing 66% faster than others, with AI accelerating across all industries. Workers with AI skills command 56% wage premiums while AI-savvy industries see 3x higher revenue per employee growth. Even "automatable" jobs see growth as AI upgrades roles rather than replacing workers entirely.

🧠💚 AI beats humans on emotional intelligence tests

University of Geneva researchers found ChatGPT and other AI systems beat humans on emotional intelligence tests, scoring 81% versus 56% for humans. Six models including GPT-4 and Claude selected emotionally appropriate responses to complex scenarios better than people. While AI can't "feel" emotions, its ability to display optimal emotional intelligence shows promise for mental health and customer service.

🏥🎯 FDA approves AI tool to predict breast cancer risk

FDA authorized Clarity Breast, the first AI platform predicting breast cancer risk from routine mammograms, enabling commercial launch this year. The AI analyzes subtle patterns invisible to humans, generating five-year risk scores without demographic data. Testing revealed half of younger women showed risk levels typical of older patients, challenging age-based screening protocols.

❤️👣 AI foot scanner predicts heart failure weeks early

Cambridge startup Heartfelt Technologies developed an AI wall scanner monitoring ankle swelling, predicting heart failure up to 13 days before emergency care. The scanner captures 1,800 images per minute measuring fluid accumulation, predicting five of six hospitalizations in NHS trials. Over 80% of participants kept the scanner post-study, showcasing proactive home health monitoring's future.

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