Weekly AI News

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

AI evolves so fast we forget to celebrate its birthdays! We missed the 8th anniversary of the "Attention is All You Need" research paper. The one that started it all. Without it, no ChatGPT, no Claude, nada.

Before this 2017 Google paper, AI was like trying to follow a conversation during a metal concert - impossible. The "transformer" architecture gave AI this magical ability to focus on what really matters in a sentence. The 8 researchers who wrote it? They've become legends.

Oh and by the way, Meta pulling out the checkbook with $100M to poach OpenAI geniuses (plot twist: nobody bites), OpenAI and Microsoft bickering like an old couple over 33% equity, and Midjourney dropping in with its acid trip-style videos.

The best part? A solo developer sells his "vibe coding" startup for 80 million. Meanwhile, Scale AI gets ghosted by everyone after Meta's investment.

Moral of the story: in 2025, AI is 50% innovation, 50% Silicon Valley telenovela. And to think we owe it all to 8 nerds and their paper on attention...

Happy reading!

🎬🤖 Midjourney launches V1, its first AI video model with surreal style

Midjourney has released V1, its long-awaited image-to-video model that transforms static images into 5-second animated clips. Available exclusively through Discord, V1 generates four videos per prompt at 8x image costs—claimed to be 25x cheaper than rivals. While not designed for realism, V1 maintains Midjourney's signature surreal aesthetic as the company aims to build real-time open-world simulation tools. Check the video

🎨✨ Midjourney integrates Style Reference in V7

The "Style Reference" feature allows users to apply reference visual styles to generated images with remarkable consistency. This functionality limits Midjourney's sometimes excessive creativity while improving moodboard coherence. Users can now better control their creations' aesthetics using the "--sref" parameter.

🪖💻 Meta, OpenAI, Palantir Tech Execs to advise U.S. Army on technology

Meta CTO Andrew Bosworth, Palantir CTO Shyam Sankar, OpenAI CPO Kevin Weil, and former OpenAI research chief Bob McGrew join a new U.S. Army innovation corps. These tech executives will serve as technical experts advising the Army's modernization efforts in their free time. The move deepens ties between major tech companies and U.S. defense agencies.

🛡️💰 OpenAI inks $200 Million deal with Pentagon for cybersecurity

The Department of Defense awarded OpenAI a $200 million contract to develop AI for national security challenges in warfighting and enterprise domains. This marks OpenAI's first defense technology contract, focusing on bolstering cybersecurity and administrative systems for service members. The one-year deal establishes OpenAI as an official Pentagon contractor with work centered in Washington, D.C.

🔥⚡ OpenAI, Microsoft partnership hits 'boiling point'

The WSJ reports the OpenAI-Microsoft partnership is at a "boiling point" over disputes about compute access, IP rights, and company restructuring. OpenAI is considering filing antitrust complaints—the "nuclear option"—against Microsoft over anticompetitive behavior. The conflict escalated over OpenAI's $3B Windsurf acquisition, with OpenAI seeking to reduce dependency on Microsoft by partnering with Google for cloud compute.

📊🤝 OpenAI wants Microsoft to accept 33% stake in business

OpenAI has proposed Microsoft accept a 33% equity stake in its restructured for-profit business, allowing the startup to eventually go public. In exchange, OpenAI wants Microsoft to relinquish its right to 49% of future profits, 20% of revenue, and IP usage rights through 2030. Microsoft is negotiating for continued access to OpenAI's intellectual property after their current commercial deal ends.

🎙️📝 OpenAI launches Record Mode meeting assistant

OpenAI introduced Record Mode, a new meeting assistant that records, transcribes, and summarizes conversations. Available on ChatGPT Pro, Team, Enterprise and Edu for Mac users, this feature aims to streamline meeting documentation. The tool integrates seamlessly with existing ChatGPT subscriptions for enhanced productivity.

🎧🚀 OpenAI launches podcast with GPT-5 summer timeline

OpenAI introduced a new podcast hosted by former engineer Andrew Mayne, where CEO Sam Altman revealed GPT-5 should "probably" arrive this summer. GPT-5 will simplify ChatGPT's confusing model lineup into one streamlined UI, ending the "whole mess" of current model choices. Altman noted users are "surprisingly willing to wait for a great answer" for complex problems, driving development of thoughtful reasoning models.

🧬⚠️ OpenAI prepares for bioweapon risks

OpenAI published new safety measures anticipating its next-generation models could reach "high risk" status for biological weapon creation. Mitigations include training models to refuse harmful requests, deploying always-on detection systems, and advanced red-teaming. The company plans a July biodefense summit with government researchers and NGOs to discuss risks and countermeasures.

💸🏢 OpenAI starts selling ChatGPT at a discount, hurting Microsoft

OpenAI has begun discounting enterprise ChatGPT subscriptions when customers commit to spending on additional AI products. This move frustrates Microsoft salespeople who sell competing apps without similar discounts. The discounting suggests OpenAI's pricing power with enterprises may not match its consumer strength, while reflecting efforts to pass savings to large customers.

📁🔍 AI watchdogs detail OpenAI concerns

The Midas Project and Tech Oversight Project launched the OpenAI Files, a comprehensive hub documenting potential conflicts of interest and governance failures at OpenAI. The report examines four areas: Restructuring, CEO Integrity, Transparency & Safety, and Conflicts of Interest. The initiative includes a "Vision for Change" proposing standards for AI firms as AGI approaches.

💬🔓 Meta AI app leaks users' private chats, exposing sensitive data

Meta's standalone AI app faces criticism for a serious privacy flaw where users unintentionally publish private conversations publicly. The app's "share" feature posts chats without clear warnings, exposing everything from legal confessions to home addresses. Critics call it a "privacy disaster" as Meta remains silent on fixing this design oversight.

🧠💼 Meta in talks to hire former GitHub CEO for AI efforts

Meta is in advanced talks to hire Nat Friedman, former GitHub CEO turned prominent AI investor, to boost its artificial intelligence efforts. The company is also discussing bringing on Daniel Gross, Friedman's partner in investment fund NFDG. These high-profile hires signal Meta's aggressive push to strengthen its AI capabilities.

✂️🤝 OpenAI severs ties with Scale AI after Meta's investment

OpenAI has dropped Scale AI as a data provider following Meta's $14.3B investment in the startup. While initially planning to continue the partnership, OpenAI now says it's shifting to more specialized partners for future frontier models. The move reflects concerns about vendor neutrality as Scale deepens ties with Meta.

🔌📊 Google reportedly cutting ties with Scale AI

Google plans to end its partnership with Scale AI after previously budgeting $200 million for 2025 services. Microsoft may follow suit, while OpenAI has already scaled back work with the startup. As Meta takes a 49% stake and names Scale's CEO to lead its superintelligence push, rival AI companies grow wary of relying on competitor-controlled infrastructure.

🎤🔍 Google tests real-time AI voice chats in Search

Google is rolling out Search Live in AI Mode, enabling back-and-forth voice conversations with its AI chatbot directly from search. Currently testing with Labs users in the US, the feature uses a custom Gemini version for real-time web searches. Google plans to add camera-sharing capabilities in coming months for visual queries.

🚀🌟 Gemini 2.5 family goes GA with new flash-lite

Google graduated its Gemini 2.5 Pro and Flash models to stable production status while launching a hyper-efficient Flash-Lite variant. The 2.5 Flash-Lite beats previous Lite models across benchmarks while maintaining a 1M token context window. All three models feature adjustable "thinking" capabilities for controlling reasoning depth and cost.

📹🎬 YouTube is plugging Veo 3 AI videos directly into Shorts

YouTube will integrate Google's Veo 3 AI video generation model into Shorts later this summer, according to CEO Neal Mohan. The upgrade brings improved video quality and audio capabilities to Shorts creators. It's unclear whether creators will need paid subscriptions to access Veo 3 features.

🔧🌐 Anthropic launches remote MCP server support

Anthropic launched remote MCP server support for Claude Code, eliminating the need for local server management. Users can now pull context from their tools directly into Claude Code. This update streamlines the development workflow for Claude Code users.

⚡🏭 xAI faces lawsuit over unpermitted gas turbines at data center

Elon Musk's xAI faces legal action for running 400+ MW of natural gas turbines without required air permits at its Memphis Colossus data center. The Southern Environmental Law Center alleges Clean Air Act violations, citing potential emissions of 2,000+ tons of nitrogen oxides annually. The case highlights environmental costs of AI infrastructure in Memphis, already ranked as the U.S. "asthma capital."

🧠🔓 MiniMax's open reasoner with 1M token context

Chinese startup MiniMax released M1, an open-source reasoning model with a 1M token context window achieving comparable performance to leading models. M1 excels in software engineering and tool use, with training costing just $535k over three weeks using their new CISPO algorithm. The efficiency demonstrates how innovative architecture challenges assumptions about AI scaling costs.

🎥🚀 MiniMax debuts Hailuo 02 video model

MiniMax launched Hailuo 02, a new AI video model that ranks No. 2 on the Artificial Analysis leaderboard, surpassing Veo 3. The 1080p generator shows significant quality improvements over its predecessor while remaining economical at approximately $0.44 for 6 seconds of HD video. The model strengthens competition in the increasingly crowded AI video generation market.

🖼️✂️ Higgsfield unveils Canvas for pixel-perfect editing

Higgsfield unveiled Canvas, an AI image editing model with pixel-perfect product placement capabilities. The tool allows direct object selection and inpainting with prompts for precise control. While less accurate than Flux Kontext on complex images, Canvas positions itself as a new standard for AI-powered product placement.

🏗️💰 Amazon to spend $13 Billion on Australia AI data centers

Amazon will invest $13 billion over four years in new Australian data centers to expand its AI business. This major investment represents a significant win for Australia and reflects web giants' determination to capitalize on global AI interest. The infrastructure will support Amazon's growing AI services in the Asia-Pacific region.

🛒🤖 China's AI avatars outsell humans in livestream

Chinese entrepreneur Luo Yonghao's AI digital twin generated over $7M in sales during a six-hour Baidu livestream, outperforming his human-led streams. The AI hosts promoted 133 products to 13M viewers, with over 100k digital humans reportedly working in China's $946B live commerce sector. AI avatars slash costs by 80% while increasing transactions by 62% on average. Check the video

👥📱 TikTok will let brands generate AI influencer content

TikTok announced new capabilities for Symphony, its AI ads platform, enabling brands to create content mimicking human influencers. Advertisers can generate videos with virtual avatars holding products, trying on clothing, and displaying apps on phone screens. The system moves beyond basic video generation to replicate authentic influencer-style content.

💵📈 Scale AI's top rival hits $1 Billion in sales

Surge AI generated over $1 billion in sales last year, surpassing Scale AI's $870 million revenue despite having no venture capital. Unlike Scale, Surge is profitable and bootstrapped, casting doubt on Meta's recent $14.3 billion investment for 49% of Scale. Surge's quiet dominance reveals Scale may not be as competitive in data labeling as previously thought.

💻💰 Solo-owned vibe coding startup sells for $80M

Developer Maor Shlomo sold his bootstrapped startup Base44 to Wix for $80M after reaching 250k users and $189k monthly profits in six months. The platform enables non-programmers to build apps with natural language, growing to 10k users in three weeks through word-of-mouth. Shlomo's eight employees receive $25M in bonuses from the acquisition.

🏢❓ McKinsey details AI investment 'paradox'

McKinsey's report reveals 80% of companies use AI but see almost no material earnings impact, identifying a "genAI paradox." The firm argues success requires rebuilding processes around AI agents rather than inserting them into existing workflows. Companies using general-purpose AI tools see improvements too diffuse to measure in financial results.

👷📊 Stanford study reveals what workers want from AI

Stanford surveyed 1,500 workers on AI automation preferences, finding 41% of YC startups focus on areas workers consider low priority. Workers primarily want to automate repetitive tasks like scheduling and data entry, with nearly half preferring equal human-AI partnership over full automation. Arts/media professionals show strongest resistance, with only 17% of creative tasks receiving positive automation ratings. Read the report

🔄🧠 MIT researchers teach AI to self-improve

MIT developed Self-Adapting LLMs (SEAL), enabling models to create their own training data and self-update instructions. The framework uses reinforcement learning, with AI learning more effectively from its own notes than GPT-4-generated materials. SEAL improved puzzle-solving from 0% to 72.5%, pointing toward a future where models upgrade exponentially on their own.

📝🧠 MIT study shows ChatGPT's detrimental impact on cognition

MIT research found students using ChatGPT for essay writing showed significantly weaker brain activity and memory retention than those writing unaided. The four-month EEG study of 54 students revealed ChatGPT users displayed weakest neural connectivity across creativity, memory, and processing regions. Results highlight concerning tradeoffs as AI tools integrate rapidly into education systems. Read the study

🧩🤖 AI develops human-like object understanding

Chinese scientists discovered AI models spontaneously develop internal 'maps' mirroring human conceptual understanding. Testing 4.7M decisions across 2,000 objects, AI developed 66 core categorization methods closely matching human brain patterns. Rather than memorizing patterns, models build genuine internal concepts, suggesting machine intelligence may work similarly to human cognition.

👶💻 UK study reveals AI's hidden impact on children

The Alan Turing Institute found 22% of UK children aged 8-12 use AI, with private school students (52%) nearly 3x more likely than state school peers (18%). Some children refuse AI after learning about environmental impacts, while most use it for creativity and learning. The digital divide threatens to widen educational gaps as this generation grows up with AI.

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