Weekly AI News

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

Well, we really didn’t see this one coming. An image generator clever enough to create menus, infographics, and glasses of wine (filled to the brim, please) without everything going haywire with every tweak. And yet, GPT-4o’s image version has proven the opposite (possibly silencing a few detractors who claim OpenAI’s days are numbered).

We can refine every detail, preserve the visual structure, and do it all by casually chatting with the tool. The catch? Even if we feel like a one-day Picasso, we don’t become true artists just by saying, “Give me a Ghibli style.”

Then there’s the legal side: the way these models are trained remains a serious gray area.

Meanwhile, Elon is playing the billionaire-collector by having xAI acquire X, like a pro at cosmic strategy. In short, it’s thrilling, exciting… and packed with thorny issues to handle.

Happy reading!

🎨🤖 OpenAI adds image generation to GPT-4o and Sora

OpenAI has unified text and image capabilities so ChatGPT can produce more precise, context-aware visuals. GPT-4o now treats images as part of its multimodal understanding, improving text clarity and graphics. Users can edit images with natural language while preserving coherence across versions. This new feature replaces DALL-E 3 as the default image generator for Free, Plus, Pro, and Team subscribers.

⚙️✨ GPT-4o’s new update for paid users

OpenAI rolled out an enhanced GPT-4o version offering better adherence to instructions, more creativity, and extra “freedom” in responses. The model now handles multiple requests in the same prompt more efficiently and solves complex coding tasks with greater ease. It also shows improved intuition, generating more inventive ideas. Finally, OpenAI has reduced the default number of emojis in replies.

💹📈 OpenAI expects revenue to triple to $12.7 billion

The company predicts astonishing growth after hitting $3.7 billion in revenue last year. With 2 million businesses already signed up for the corporate version of ChatGPT, OpenAI is testing new premium plans, including a $200/month Pro tier. It may also roll out high-end services costing thousands per month. Despite the surge in revenues, OpenAI doesn’t foresee being cash-flow positive before 2029.

💰🚀 OpenAI nears a massive $40 billion funding round

Reports suggest SoftBank will lead this historic financing, boosting OpenAI’s valuation to $300 billion. SoftBank plans an initial $7.5 billion, followed by $22.5 billion with other backers such as Magnetar Capital and Founders Fund. OpenAI anticipates tripling revenue to $12.7 billion by 2025 and projects profitability by 2029 with $125 billion in predicted revenue. The deal also supports Stargate, the $300 billion AI infrastructure venture launched with SoftBank and Oracle.

🇮🇳💸 OpenAI considers cutting ChatGPT’s price in India

OpenAI is discussing lowering ChatGPT’s monthly subscription cost by 75–85 % in a market where $20 can be relatively expensive. The company aims to hit 1 billion daily active users by year’s end, making growth in India vital. Talks are underway with Reliance Industries, led by Mukesh Ambani, to distribute ChatGPT or sell OpenAI’s models via API. Reliance’s Jio carrier might be a key channel to extend ChatGPT’s reach.

🏗️📂 OpenAI mulls building its own data center

The company is reportedly weighing a huge investment in storage hardware and software, a move that would make it among the world’s largest storage customers. By seeking up to 5 exabytes of storage, OpenAI aims to lessen its dependence on cloud partners like Microsoft or Oracle. This approach could give it tighter control of the critical data needed to train its AI models.

⚖️📰 Judge allows key copyright claims against OpenAI

A New York federal judge allowed most of the New York Times’ copyright infringement case to proceed, rejecting OpenAI’s motion to dismiss. The Times alleges OpenAI used the paper’s content without compensation to train ChatGPT, citing passages that appeared to be lifted verbatim. OpenAI argues using publicly accessible data is lawful, but the court will now determine whether the Times’ infringement claims hold. The trial date is not yet set.

🏷️🤝 OpenAI adopts Anthropic’s MCP protocol

CEO Sam Altman announced that OpenAI will integrate Anthropic’s Model Context Protocol (MCP) into its products, including ChatGPT and the Agents SDK. MCP is an open-source standard that connects AI systems to various data sources and tools to generate more accurate responses. Anthropic welcomes OpenAI’s support, highlighting its own platform’s success integrating MCP. OpenAI plans to share more details on MCP soon.

🔎🧠 Anthropic reveals how Claude “thinks”

Two new research papers shed light on Claude’s internal mechanisms, showing how it uses a “language of thought” across multiple languages and plans rhymes in advance when writing poetry. Claude refrains from speculation unless sufficiently confident, reducing hallucinations. The research offers insight into how the model processes inputs step by step—a crucial step as we inch closer to more powerful AI systems.

 🎼⚖️ Anthropic wins a round in music publishers’ lawsuit

A federal judge in California denied music publishers’ request to block Claude from training on copyrighted song lyrics. Plaintiffs like Universal Music Group cited at least 500 songs by Beyoncé and the Rolling Stones. The judge said the injunction was overly broad and did not show irreparable harm. The question of fair use remains unresolved, though it will likely be pivotal in upcoming court proceedings.

🎯🤳 Perplexity’s bold move to acquire TikTok’s U.S. operations

The AI search startup offers to rebuild TikTok’s recommendation engine with American oversight while embedding its search tech. Perplexity promises full transparency, open-sourcing the algorithm on U.S. servers and boosting performance with Nvidia Dynamo for 100x scale. TikTok videos would appear in Perplexity’s AI search, and Perplexity’s system would power TikTok’s in-app search. While it might be a publicity stunt, the looming ban deadline suggests we’ll know soon.

🏆🤔 Google’s Gemini 2.5 Pro tops the AI leaderboard

Google unveiled Gemini 2.5 Pro Experimental, claiming the #1 spot on LMArena and boasting advanced reasoning in math, science, and coding. With coding scores of 63.8 % on SWE-Bench Verified and 68.6 % on Aider Polyglot, it excels particularly in web apps. A 1 million-token context window comes standard, with a plan to expand to 2 million tokens. Available in Google AI Studio and the Gemini app for Advanced subscribers, its lead could still be challenged by GPT-5 or others.

📱👀 Google rolls out ‘Project Astra’ for real-time vision in Gemini

Google One AI Premium users can now let Gemini “see” their phone screen or camera feed in real time. One user on Reddit demoed Gemini’s ability to read and interpret on-screen text or images instantly. This major step follows nearly a year of teasers about the Astra project. It could enable interactive visual assistance, from scanning documents to identifying objects in real-world settings.

📄🔒 Google removes DEI terms from some internal studies

Google employees have been instructed to avoid language related to diversity, equity, and inclusion in certain product research. Words like “equity” and “inclusion” must be replaced with phrases like “build for all,” and staff have been told to retroactively edit older documents to remove DEI references. This directive suggests a cautious approach to politically sensitive terms in internal research.

💬🤖 Meta tests AI-generated comments on Instagram

Instagram is experimenting with a “Write with Meta AI” feature that suggests comments after analyzing the photo’s content. This prototype follows a similar test on Facebook. Critics say AI-generated responses compromise authentic engagement, although Meta is deepening its platform-wide integration of AI. No timeline for a public release has been announced.

💡💻 Microsoft explores compensating AI training contributors

Led by Jaron Lanier, Microsoft’s project examines “training-time provenance,” which aims to track how individual data points shape AI outputs. This move could pave the way for creators to be recognized or paid when their content significantly informs AI generations. With lawsuits claiming AI labs violated copyrights, and some companies offering partial royalties, Microsoft’s research might spark broader adoption of data compensation models.

🌌🔇 DeepSeek unveils a quiet V3 upgrade

The Chinese AI startup launched V3-0324, a 641 GB model featuring a Mixture-of-Experts design that only activates 37 billion parameters per token. Early tests confirm it can run on Apple’s Mac Studio—a rarity for a model this size. Licensing has switched to a permissive MIT model, fueling excitement for the rumored R2 version, which might shake the AI landscape once again.

🦾💡 Tencent unveils Hunyuan T1 reasoning model

Hunyuan T1 rivals DeepSeek’s R1 and OpenAI’s 4.5 in performance, adopting the industry’s first hybrid Transformer-Mamba architecture for faster inference. It outperforms or equals top competitors in benchmarks for math and Chinese comprehension. Tencent matches DeepSeek’s pricing at roughly $0.14 per million input tokens and $0.55 per million output tokens. As competition grows among Chinese tech giants, the global AI lead may be up for grabs.

👁️🧮 QVQ-Max, Alibaba’s new visual reasoning model

Alibaba’s Qwen team introduced QVQ-Max, which goes beyond standard image recognition to tackle geometry problems, code generation, and creative projects. The model uses adjustable “thinking time,” improving accuracy the longer it processes. Demos include blueprint analysis and user sketch refinement. Qwen plans to develop a fully autonomous visual agent for gaming and device operation in the near future.

🔊📸 Alibaba’s all-in-one AI for mobile

Alibaba released Qwen2.5-Omni-7B, a multimodal model handling text, images, audio, and video in real time, lightweight enough to run on smartphones. It employs “Thinker-Talker” for seamless transitions among different data modalities, outperforming specialized audio solutions in benchmarks. Available open source on Hugging Face and GitHub, it paves the way for practical AI agents with wide-ranging applications.

🚗🤖 BMW and Alibaba partner to bring AI to cars

The two firms announced a collaboration to develop an in-car AI assistant, leveraging Alibaba’s Qwen for advanced voice recognition and contextual understanding. Users will get real-time info on dining, parking, and traffic, all triggered by voice commands instead of on-screen interfaces. BMW will also roll out Car Genius for maintenance and Travel Companion for planning. Multimodal inputs such as gesture and eye tracking promise a more intuitive driving experience.

🎨🚀 Ideogram launches its 3.0 image model

The startup’s new version promises major leaps in photorealism, accurate text rendering, and style control. Ideogram 3.0 surpasses leading text-to-image generators like Imagen 3 and Recraft V3, based on user tests. Enhanced layout and logo creation are possible, and users can upload up to three style references for consistent results. All features remain free, although the debut competes with the hype around OpenAI’s 4o.

🌙🔥 Reve’s stealthy image model claim

Reve emerged with Reve Image 1.0, codenamed “Halfmoon,” which recently soared to #1 on the Image Arena leaderboard. It boasts extraordinary adherence to prompts, excellent text generation, and photorealistic quality, rivaling Imagen 3 and Midjourney v6.1. Features include natural language editing and an explore tab for community prompts. Although the API isn’t live, the free preview already attracts significant attention for bridging top-tier visual fidelity and text capability.

🤫🧩 Pika’s secret new feature for video object manipulation

Rumors suggest Pika is developing a tool that isolates and edits any character or object within a video while leaving the rest untouched. This tech could swap actors between scenes, change outfits, or adjust animations with unprecedented precision. Although details remain scarce, insiders expect a release within a few months, potentially transforming video editing workflows.

⌚📷 Apple Watch might get cameras in an AI push

Apple is experimenting with camera hardware and AI functionalities for its smartwatch, potentially enabling real-time environmental awareness. The plan mirrors broader “Visual Intelligence” efforts, which may also revamp AirPods with vision-based features. While AI hardware initiatives have often struggled for mainstream success, Apple’s massive ecosystem could give it a unique edge—unlike Humane, which was recently acquired by HP after failing to scale.

🏦🖥️ Apple’s $1 billion bet on Nvidia servers

Reports say Apple is purchasing around 250 Nvidia GB300 NVL72 servers for $3.7–4 million each, enlisting Dell and Super Micro to build large-scale AI infrastructure. This marks a strategic pivot as Apple races to catch up in generative AI, given Siri’s development setbacks. Though Apple has explored proprietary AI chips, these public investments signal the need for external computing muscle to compete effectively.

🏢🤝 PwC’s “agent OS” unifies AI for the enterprise

PwC introduced a central hub for orchestrating AI agents across various platforms, promising to deploy and scale them up to 10x faster. The system integrates with major providers like Anthropic, AWS, Microsoft, and Salesforce, simplifying workflows. Use cases include faster customer support and brand compliance. PwC itself runs over 250 internal AI agents, showcasing its commitment to large-scale, multi-agent environments.

🕴️⚡ xAI merges with X (formerly Twitter) in a $33 billion deal

Elon Musk finalized an all-stock acquisition combining his social media platform with his AI firm xAI. The new valuation stands at $80 billion for xAI and $33 billion for X. Musk envisions using X’s 600 million users to train Grok, xAI’s model, offering smarter experiences at scale. He frames this merger as a milestone for “accelerating human progress” by integrating advanced AI research with a global communication network.

🎙️💼 Andreessen and Sequoia eye an investment in voice AI startup Sesame

Sesame, developing voice assistants and wearable devices, is in discussions to raise $200 million or more. Investors like Sequoia Capital, Andreessen Horowitz, and Northstar.vc are reportedly interested. While terms aren’t final, the company hinted at a possible multibillion-dollar valuation. Demand for cutting-edge voice AI is rising, fueling interest in the startup’s technology for healthcare, education, and personal assistance.

🏭🔥 Nvidia-backed cloud provider secures $225 million to buy more AI chips

Crusoe, financed partly by Nvidia, raised a $225 million debt round to purchase additional Hopper servers. It will rent them out to AI-cloud customers, betting on the lasting value of Nvidia’s hardware. Crusoe also struck a deal for up to 4.5 gigawatts of energy, enough to power millions of GPU units for data center clients like OpenAI. This move aims to position Crusoe as a key infrastructure player in the AI gold rush.

🤖📊 Databricks and Anthropic form a five-year AI partnership

Both firms signed an agreement to jointly sell each other’s AI products. Databricks customers can now leverage Anthropic’s Claude models on top of Databricks-managed data. The deal simplifies the creation of AI agents that perform tasks like appointment scheduling. Databricks also announced it surpassed $3 billion in annual recurring revenue—an impressive 60% year-over-year growth.

👗🖥️ H&M embraces AI-based “digital twins”

The fashion giant is collaborating with 30 models and their agencies to create AI-generated avatars used in campaigns. These virtual models are made from multiple photos, allowing for flexible styling and scene adjustments without a traditional photo shoot. The human models retain ownership of their digital selves and earn usage-based payments. This could revolutionize how brands produce content, cutting back on costly in-person sessions.

🏆🔬 ARC Prize returns with ARC-AGI-2

The ARC Prize Foundation launched a new benchmark and a $1 million competition to foster advanced AI reasoning. ARC-AGI-2 poses tasks too difficult for current models but simple for humans, with top models scoring a paltry 4 % success rate. The foundation also added an efficiency metric, assessing cost per task alongside raw accuracy. A prize of $700k goes to whoever reaches 85 % under specified efficiency limits, nudging AI research closer to real AGI.

🏥📚 Bill Gates predicts AI replacing doctors and teachers within 10 years

Gates foresees free, high-quality medical and educational services provided by AI, drastically reducing human involvement. He coins the term “free intelligence,” describing how AI integrated into daily life could democratize specialized expertise. While acknowledging big disruptions to job markets, Gates remains optimistic about major breakthroughs in health, climate, and education—areas still requiring human-centric interaction.

🩺🔬 New AI achieves near-perfect cancer detection

Researchers introduced ECgMLP, capable of diagnosing endometrial cancer with 99.26% accuracy by analyzing microscopic slides. That’s far above the 78–81% range for human experts. It also showed high efficacy on other cancers like colorectal and breast. Such breakthroughs could transform early detection worldwide, offering precise diagnostics even in areas lacking skilled pathologists.

🤖🏠 Robots arrive in your living room

Tesla plans to produce between 10,000 and 12,000 Optimus robots this year, aiming for 50,000 next year, laying the groundwork for its first “legion” in 2025. Meanwhile, startup 1X expects to deploy its humanoid Neo Gamma in hundreds of households by late 2025. Initially, these units will be teleoperated for data collection to train AI, eventually transitioning toward more autonomous functions.

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