Weekly AI News #27

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

Last week's rumor mill says GPT-5 is dropping in August (well, maybe) after playing hide-and-seek under the codenames Starfish and Nectarine in the testing arenas.

Meanwhile, Altman's worried that young people can't decide what to eat without ChatGPT. Dude, you created a monster that swallows 2.5 billion questions a day and now you're worried we're becoming too dependent? That's like Zuckerberg discovering social media is addictive.

On the geopolitical front, Trump's rolling out his "America First" AI plan: 90 points to transform the USA into an artificial intelligence superpower. Spoiler: it requires so much electricity that the American grid, frozen since the 70s, might just blackout. Meanwhile, China's quietly building power plants...

And Google? Still second place with a measly 8.7% market share, but their infiltration strategy is genius: putting Gemini everywhere, like parsley in French cuisine. Result: we click 50% less on links when AI summarizes everything. RIP the web we knew.

Absurd bonus: a hacker modified Amazon's assistant to delete customer data, just "to see what happens." AWS fixed it without saying a word. Cybersecurity 2025 is magnificent!

Happy reading!

🔍📊 AI Market Maturity: OpenAI Dominates While Competition Intensifies

Similarweb's latest report shows OpenAI's ChatGPT maintaining 78% of AI platform traffic with 110 million daily visits, while the overall market exploded 37% in five months to reach 275 million daily interactions. Despite this dominance, serious competitors are gaining ground, Google Gemini grew 63% and Claude 26%, steadily eroding OpenAI's market share.

Google's strategy of integrating Gemini across Search, Gmail, and Drive represents a paradigm shift, transforming AI from a standalone tool into the DNA of its entire ecosystem. This free vertical integration poses an existential challenge to standalone competitors: how do you compete with omnipresence?

Surprisingly, traditional sectors show resilience. Search engines lose only 2% of traffic, design platforms thrive with 20% growth, revealing that AI complements more than replaces. Meanwhile, specialized verticals quietly explode—travel AI (Mindtrip +131%), browser automation, and voice generation (Speechmatics +124%)—while first-generation AI writing tools collapse (-14%), cannibalized by more powerful generalist models.

🚀🧠 OpenAI Readies GPT-5 for August Launch

OpenAI plans to launch GPT-5 in August, combining language capabilities with o3-style reasoning in one unified system. CEO Sam Altman described testing as a "here it is moment," claiming it instantly solved problems that made him feel "useless relative to the AI." The company also plans to release its first open-weight model since 2019 by end of July.

🏅🔢 OpenAI's Math Gold Medal Claims Under Scrutiny

OpenAI claimed its experimental model achieved gold medal-level performance at the 2025 International Mathematical Olympiad, solving 5 of 6 problems. However, Google DeepMind suggested it only earned silver, and the IMO found the premature announcement "super rude" as they'd asked AI companies to wait. The achievement remains unverified by official IMO evaluation.

💾⚡ OpenAI Expands Oracle Data Center Deal 9x to 4.5GW

OpenAI and Oracle agreed to develop 4.5 gigawatts of additional U.S. data center capacity, increasing their agreement ninefold. The deal would consume as much power as Microsoft's entire cloud infrastructure today. Oracle disclosed the partnership could generate $30 billion in revenue by 2028 alone.

🤝💔 OpenAI-SoftBank Data Center Project Faces Early Friction

The $500 billion Stargate venture between OpenAI and SoftBank has scaled back initial plans due to disagreements, potentially developing only a small data center by year-end. OpenAI has moved ahead independently, securing computing power through deals with Oracle, Google, and CoreWeave while the companies assess sites and reimagine data center design.

☁️🤖 Google Cloud Lands OpenAI as Surprise Customer

Despite being competitors, OpenAI has joined Google Cloud, giving it access to Google's infrastructure and chips potentially to train models that could erode Google's search dominance. The strategic partnership highlights how cloud providers are capitalizing on the AI boom, with Google Cloud revenue surging 32% year-over-year last quarter.

👗🔍 Google Launches AI Virtual Try-On for Shopping

Google officially launched an AI feature allowing users to virtually try on clothes using full-body photos, now available in U.S. Search, Shopping, and product results. The tool lets users see how items look on their own body rather than just models, alongside new Doppl styling app testing and smarter price alerts.

📸🎬 YouTube Adds AI Video Creation Tools for Shorts

YouTube launched new AI tools for Shorts creators including photo-to-video capabilities and Effects for quick transformations, both powered by Veo 2. Google also rolled out AI features in Google Photos with photo-to-video transformation and a new Remix editing tool for enhanced creative options.

🔗📉 Google Users Click 50% Less with AI Summaries

Pew Research found Google users click on result links 50% less when browsing pages with AI-generated summaries. Among 900 U.S. adults tracked, 58% encountered AI summaries in searches, with users rarely clicking on cited sources within the AI-generated content. Read the report

🏛️📜 Google AI Decodes Ancient Roman Inscriptions

Google DeepMind launched Aeneas, an AI system that helps historians restore and decipher damaged Latin inscriptions across the Roman Empire. The tool attributes inscriptions to provinces with 72% accuracy, dates them within 13 years, and restores damaged text at 73% accuracy, with historians finding suggestions helpful in 90% of cases.

💼🤖 Microsoft Maps Real-World AI Usage Patterns

Microsoft analyzed 200,000 Bing Copilot conversations to reveal how workers actually use AI, finding highest impact for computer science, office support, sales, and media roles. Jobs with hands-on tasks like surgeons and maintenance workers showed low AI exposure, with weak correlation between wages and AI impact contrary to predictions.

🌍💨 Mistral AI Reveals Shocking Environmental Cost of LLMs

Mistral AI published the first comprehensive environmental impact study of a large language model, revealing 20,400 tons of CO₂ emissions and 281,000 cubic meters of water consumption just from training. A single chatbot response carries 1.14g of CO₂ and 45ml of water footprint, with the company calling for mandatory environmental reporting standards. Read the article

👶🤖 Musk Plans 'Baby Grok' AI for Children

Elon Musk announced xAI will launch "Baby Grok," an AI targeted at children, though he provided no further details. The announcement comes as xAI's main product Grok has both outperformed rivals and attracted controversy, including recent incidents where it praised Adolf Hitler and called itself "MechaHitler."

📹🤖 X Plans to Revive Vine "in AI Form"

Elon Musk posted that X is planning to resurrect the beloved video app Vine "but in AI form," with the intellectual property currently owned by Twitter (now X). No additional details were provided about how AI would be integrated into the revival of the short-form video platform.

🧠🏆 Meta Hires Google's Math Olympiad AI Researchers

Meta hired three AI researchers from Google DeepMind—Tianhe Yu, Cosmo Du, and Weiyue Wang—who worked on the Gemini model that achieved gold medal-level performance at the International Math Olympiad. The hires continue Mark Zuckerberg's aggressive AI recruiting push to recover from earlier stumbles this year.

🏆🔓 Alibaba's Qwen3 Takes Open-Source Crown

Alibaba's updated Qwen3 model beats Kimi K2 across benchmarks and challenges closed-source models like Claude Opus 4. The non-thinking version activates 22B of 235B parameters with 256K-context window, achieving significant performance gains while being 100% open-source and available free on Qwen Chat.

💻🚀 Alibaba Launches Open-Source AI Coding Model

Alibaba released Qwen3-Coder, its most advanced AI coding model, alongside open-source coding agent Qwen Code. The launch intensifies competition with U.S. firms like Anthropic, joining Chinese competitors Moonshot AI's Kimi K2 and ByteDance's Seed-Coder in the rapidly evolving AI coding market.

🌐📝 Alibaba Releases Multilingual AI Translation Model

Alibaba released Qwen3-MT, an AI translation model supporting 92+ languages with strong performance across benchmarks. The model adds to Alibaba's growing suite of open-source AI tools aimed at democratizing access to advanced language technologies globally.

🧠🔄 Brain-Inspired AI Beats Giants with Just 27M Parameters

Sapient Intelligence's Hierarchical Reasoning Model uses brain-inspired architecture to beat Claude 3.7, DeepSeek R1, and o3-mini on complex tasks with only 27M parameters. Using hierarchical processing and recurrent connectivity, it achieves efficient intelligence without massive training data, already helping with rare-disease diagnostics and climate forecasting.

🔒💬 Proton Launches Privacy-First AI Chatbot Lumo

Proton launched Lumo, a chatbot with zero-access encryption where all data stays local and messages are encrypted asymmetrically. Using only open-source models like Mistral Nemo and Small 3, it offers 25 free queries weekly without account, with paid plans for more features and a "ghost mode" that deletes conversations on close.

🌐🤖 AI Browsers Add One-Click Task Automation

Dia launched an official skill gallery for smart prompts like "find events near me" as shortcut actions, while Perplexity's Comet will introduce customizable task shortcuts and natural language scripts. These features move AI browsers closer to delivering on promises of full task automation for everyday use cases.

😢🚀 Windsurf's Emotional Collapse Before Cognition Rescue

Windsurf CEO Jeff Wang revealed the startup's dramatic turnaround after OpenAI walked away and executives joined Google, leaving employees "in tears" during a Friday meeting. That same night, Cognition reached out, sparking weekend negotiations that saved the company through acquisition and Google licensing its IP.

🧠💊 Ash Raises $93M for AI-Powered Therapy Platform

Founded by Casper's Neil Parikh, Ash uses AI trained on actual therapy sessions with techniques like CBT and DBT. Unlike typical chatbots, it challenges users rather than always validating, uses reinforcement learning to track real progress, and encourages building human support networks alongside AI assistance.

📱⚠️ iOS 26 Beta Returns AI News Summaries with Warnings

Apple's fourth iOS 26 developer beta brings back AI-generated news summaries with a warning: "Summarization may change the meaning of the original headlines." This follows backlash over inaccurate summaries, including one misreporting a murder suspect's death, as Apple refines AI features without declaring a chatbot war.

💬🤖 Slack Launches AI Features for Meeting Notes and Translations

Slack announced four AI features for paid subscribers including enterprise search across databases, AI-generated meeting notes from huddles, thread summaries for mid-conversation joiners, and real-time translations. Additional features coming include hover explanations for company jargon and customized action items in activity feeds.

✈️💰 Delta Uses AI to Set Personalized Ticket Prices

Delta Air Lines is using AI to determine individualized ticket prices based on each customer's willingness to pay, currently for 3% of tickets with plans to expand to 20% by year-end. The Fetcherr technology creates real-time custom pricing, with critics calling it "predatory pricing" designed to extract maximum revenue from customers.

⌚🎙️ Amazon Acquires AI Wearable Startup Bee

Amazon is acquiring Bee, maker of a $50 AI wristband that records conversations throughout the day for summaries, insights, and reminders. The Fitbit-style device transcribes conversations and generates daily digests and to-do lists, with all employees receiving job offers though acquisition terms weren't disclosed.

🚨💻 Hacker Exploits Amazon's AI Tool to Delete Customer Data

A hacker added malicious code to AWS's Q Developer AI coding assistant that ordered it to delete customer data, exposing cybersecurity vulnerabilities in rapid AI development. While Amazon fixed the issue with no data lost, they didn't issue a public advisory, raising questions about whether security practices are keeping pace with AI deployment.

🛒🤖 Walmart Hires Instacart Exec to Lead AI Initiatives

Walmart hired Instacart's chief product officer Daniel Danker as EVP of AI acceleration, product and design, reporting directly to CEO Doug McMillon. Danker, who previously worked at Uber and Facebook, will start by August 18 as Walmart staffs up to navigate the rise of AI search and shopping tools.

🇺🇸🚀 White House AI Action Plan: America's Blueprint for Global AI Dominance

The White House unveiled a 90-point AI Action Plan to establish U.S. dominance through deregulation and massive infrastructure expansion. Three pillars: accelerate innovation by cutting red tape, build infrastructure including critical power grid expansion, and lead international diplomacy while countering China.

The infrastructure challenge is stark—America's power grid hasn't grown since the 1970s while China builds aggressively. Without expansion, the U.S. can't power competitive AI systems. Plans include streamlined semiconductor plant permits and domestic chip production.

The plan reviews FTC investigations to avoid "burdening innovation" and requires government AI contractors ensure "objective" models free from "ideological bias." For Americans, this means faster AI tools, infrastructure jobs, and better government services—though critics say it favors tech giants over public safeguards.

🏛️⚖️ White House Plans Executive Order Against 'Woke' AI

The White House is preparing an executive order requiring AI companies with federal contracts to ensure their models are politically neutral, addressing concerns about liberal bias. Expected next week, the order reflects the administration's years-long accusations of tech companies censoring conservative voices online.

🐛🤖 AI-Generated "Slop" Floods Bug Bounty Platforms

Cybersecurity teams face a new spam problem: LLM-generated bug bounty reports describing non-existent vulnerabilities that appear technically sound until examined. Platforms like HackerOne and Bugcrowd now use their own AI tools to detect hallucinated vulnerabilities, creating an arms race between AI attackers and defenders.

🎮🧩 ARC Launches Interactive AGI Test Stumping Frontier Models

ARC Prize released ARC-AGI-3 preview featuring three original games testing AI agents' ability to generalize through trial and error with no instructions. Early results show frontier models like OpenAI's o3 and Grok 4 struggle with basic levels that humans find easy, pushing research toward true artificial general intelligence.

🧠🎭 AI Models Fall for Human Psychological Tricks

Wharton research shows AI models including GPT-4o-mini can be manipulated using Cialdini's principles of influence like authority and scarcity, doubling compliance to objectionable queries from 33% to 72%. The findings reveal AI models share human-like vulnerabilities to psychological persuasion tactics, requiring new defensive approaches.

🔬🧬 AI Models Transmit Hidden Learning Traits

Anthropic researchers discovered "subliminal learning" where teacher models transmit preferences or behaviors through unrelated data to student models sharing the same architecture. Models trained on code from an owl-loving teacher developed owl preferences despite no animal references, with effects extending to dangerous behaviors even through filtered content.

🦠💊 AI Designs Cancer-Killing Proteins in Weeks

Danish scientists developed an AI platform designing custom "minibinder" proteins in weeks rather than years, enabling T cells to target cancers like melanoma. Using AlphaFold2 and virtual safety screening, the system creates both common and patient-specific treatments, demonstrating AI's potential for personalized medicine breakthroughs.

🏥🇰🇪 OpenAI's Medical Copilot Reduces Errors in Kenya

OpenAI partnered with Penda Health in Nairobi, finding AI copilots reduced diagnostic errors by 16% and treatment mistakes by 13% across 40K patient visits. The AI Consult system monitors decisions in real-time as a safety net, with 75% of clinicians reporting substantial quality improvements through capable models, seamless integration, and active training.

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