Weekly AI News

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

The world of artificial intelligence (AI) is buzzing with major developments this week! SoftBank is making a bold move by injecting $3 billion annually into OpenAI’s technology to launch “Cristal Intelligence”, while Google unveils the Gemini 2.0 lineup, featuring a Pro model dedicated to advanced programming. Open-source projects are also gaining traction, as seen with Zyphra’s rise.

At the same time, the European Union is tightening regulations by banning “unacceptable risk” AI applications, and ByteDance is pushing the boundaries of realism with its latest AI-powered avatar system. These groundbreaking announcements come amid growing concerns over AI’s rapid advancement.

As Paris hosts the Summit for Action on Artificial Intelligence on February 10–11, 2025, bringing together nearly a hundred countries and key industry players, discussions are centered on digital sovereignty, transparency, and security challenges. AI is more than ever at the heart of global debates.

🇯🇵🤝 SoftBank and OpenAI Announce ‘Cristal Intelligence’

SoftBank commits to a yearly $3B investment in OpenAI’s tech while launching Cristal Intelligence, delivering customised enterprise ChatGPT tools exclusively in Japan. The venture offers secure integrations and could rival established consulting firms. It follows SoftBank’s “Stargate” data center partnership with OpenAI and Oracle, signaling a significant Japanese push into the global AI race.

🧠📚 OpenAI Debuts AI Research Assistant

OpenAI introduces “Deep Research,” a ChatGPT feature that scours the web, images, and PDFs for in-depth reports with citations in under 30 minutes. Initially, for Pro subscribers at $200/month, it tackles complex, multi-step questions that can take humans days to solve. Powered by a specialised o3 model, it outperforms competing AIs like Gemini Thinking on “Humanity’s Last Exam.” This marks a shift toward more autonomous, long-form AI reasoning within ChatGPT.

💬🔓 Sam Altman’s Stance on Open Source

During a Reddit AMA, OpenAI CEO Sam Altman admitted the company may have been “on the wrong side of history” regarding open-source AI. He hinted the full o3 model might arrive “sooner than a few months,” while also praising the rival DeepSeek model. CPO Kevin Weil revealed more agents and a new image generator are coming soon. The biggest takeaway is Altman’s open-source comments, which could signal a major philosophy change at OpenAI.

📈💸 ChatGPT Subscribers Nearly Tripled; OpenAI Discusses Open Sourcing Models

OpenAI’s ChatGPT jumped from 5.8 million paying users to 15.5 million, pushing estimated annual revenue beyond $4 billion. Its API usage skyrocketed, too, possibly generating $3.2 billion a year. However, Chinese competitor DeepSeek offers cheaper open-source models, posing a clear threat. CEO Sam Altman responded by saying OpenAI might release some model weights, reflecting a shifting stance toward open source.

🤔🔗 OpenAI Updates o3-Mini with Transparent Reasoning

Just days after launching o3-Mini, OpenAI now displays a partial “chain of thought” for answers, revealing how the model arrives at conclusions. Free and paid users alike gain this feature, though it’s not as detailed as DeepSeek’s R1. This is the first time OpenAI’s advanced reasoning is available on the free tier, albeit with strict rate limits. The change underscores OpenAI’s pivot toward more transparency.

🤖📲 OpenAI Signals Hardware Push with Trademark Filing

OpenAI filed a broad trademark covering humanoid robots, AI wearables, smartwatches, and more. CEO Sam Altman has spoken of teaming up with Jony Ive on an “AI-first phone” and rebuilding the robotics division. The filing spans devices for “AI-assisted interaction,” hinting at a consumer hardware play. It suggests OpenAI may challenge established tech giants in a new arena.

🎨🔧 OpenAI Debuts First-Ever Rebrand

OpenAI has refreshed its brand identity, unifying previously mismatched fonts, colors, and logos. The new look aims for a human, organic feel while projecting research-driven authority. From ChatGPT to the website interface, every product sports a cohesive design. This is OpenAI’s first major rebrand and reflects its growing influence across consumer and enterprise markets.

🔥🏈 OpenAI Makes Super Bowl Debut with “The Intelligence Age”

OpenAI aired a 60-second spot during Super Bowl 59, positioning ChatGPT alongside humanity’s greatest inventions, from fire to television. The ad shows how ChatGPT can handle everyday tasks, visually presented with a pointillism-inspired style mixing 2D and 3D animation. Despite AI being the core subject, the final production was crafted by human creatives. This marks OpenAI’s major push into mainstream marketing, aiming to cement ChatGPT’s status as an everyday tool.

🎓🏫 OpenAI Partners with California University System

OpenAI will roll out ChatGPT Edu to 460,000 students and 63,000 staff across 23 campuses in California. The deal includes curriculum integration, research support, and data privacy protocols. Officials emphasise responsible use and academic integrity in this pioneering large-scale AI deployment. It’s a milestone in embedding generative AI across higher education.

🇪🇺🔒 OpenAI Launches Data Residency in Europe

OpenAI now offers European data residency, helping local organisations comply with GDPR, Germany’s Federal Data Protection Act, and other privacy regulations. Eligible API endpoints, plus new ChatGPT Enterprise and Edu accounts, can store data at rest in Europe. This mirrors moves by other tech giants to meet sovereign data requirements. The initiative affirms OpenAI’s commitment to privacy in EU markets.

🖼️✨ ChatGPT Integrates Finegrain for Advanced Image Editing

OpenAI adds Finegrain, enabling users to edit images directly within ChatGPT via natural prompts. The tool can achieve near-Photoshop-level retouching and enhancements. This expansion aims to simplify image manipulation for everyday users, eliminating the need for specialised software. ChatGPT has become more of an all-in-one creative and conversational assistant.

⚡💻 Google Rolls Out Gemini 2.0 Lineup with Pro

Google’s Gemini 2.0 lineup includes the Pro Experimental model with a massive 2M-token window and powerful coding capabilities, plus budget-friendly Flash and Flash Lite. The 2.0 Flash Thinking reasoning model is freely available in the Gemini app with real-time chain-of-thought insights. Despite the hype, early benchmarks trail top-tier competition from OpenAI. All models feature multimodal support, setting the stage for broader future releases.

🖌️📷 Imagen 3 Arrives in the Gemini API

Google’s Imagen 3 can now be accessed through the Gemini API, initially for paying customers. This latest text-to-image model creates high-fidelity visuals in a range of styles, from photorealism to anime. At $0.03 per image, developers get control over aspect ratios, batch generation, and more. A free-tier rollout is planned, promising advanced image generation for all.

🔎🤖 Google Starts Testing New Search ‘AI Mode’ Internally

Google employees are trialing an “AI Mode” in Search that provides AI-generated overviews for open-ended queries. Powered by Gemini 2.0, it targets complex, exploratory questions beyond traditional keyword queries. Early interface leaks show an AI Overview panel integrated into search results. If successful, this could fundamentally change how users interact with Google Search.

📸💧 Google Adds AI Watermarks to Magic Editor Images

Google Photos now applies SynthID watermarks to images altered by Magic Editor’s generative AI. The system embeds metadata to flag AI-generated or AI-edited content. While Google acknowledges that tiny changes may slip through, this marks a step toward transparency in image manipulation. Similar watermarking efforts, like Adobe’s Content Credentials, are emerging industry-wide.

🛑🔍 Google Ends Its Promise Not to Use AI for Weapons, Surveillance

In a controversial pivot, Google dropped its explicit ban on using AI for weapons or surveillance. The new policy commits to “responsible AI” with oversight but raises internal and external concerns. Employee protests highlight potential ethical conflicts, especially with military contracts. The shift mirrors a similar move by OpenAI last year, spotlighting the complex ethics around AI in defense.

🤯🎥 ByteDance’s AI-Human Avatar Model Breakthrough

TikTok parent ByteDance unveiled OmniHuman-1, a deepfake system generating ultra-realistic videos from a single reference image and audio. It can adjust body proportions, handle wild poses, and maintain style continuity. Despite legal crackdowns on AI impersonation in several U.S. states, regulation and detection remain difficult. This technology signals a new era of seamless, high-fidelity deepfakes.

🆓🌐 Hugging Face Released open-Deep-Research

Hugging Face launched open-Deep-Research, an open-source challenger to OpenAI’s Deep Research. It scores 55% on GAIA benchmarks and features autonomous web navigation for extensive data gathering. The platform invites community contributions to refine and expand its capabilities. This move underlines Hugging Face’s commitment to democratising cutting-edge AI research.

🚀💰 Open-Source AI Startup Zyphra Targets $1 Billion Valuation

Zyphra, known for small-footprint models that run on everyday devices, is in talks for a $100M Series A at a $1B valuation. Investors like Metaplanet and Future Ventures see promise in Zyphra’s low-parameter approach, which lowers costs. Despite having no revenue yet, the company aims for $30M annualised revenue by year-end. Amid competition from DeepSeek, Zyphra bets on specialized edge-friendly AI.s

🔥🏆 Ai2’s Tülu3-405B Surpasses DeepSeek and OpenAI

Ai2’s Tülu3-405B model outperforms rivals like DeepSeek with “reinforcement learning from verifiable rewards” (RLVR). This method penalises or rewards accuracy on objectively checkable tasks, boosting reliability. Tülu3-405B’s success highlights a new milestone in the global “AI race.” Refining answers post-training sets a bar for models that can consistently handle tough math and instructions.

🔐🎯 Anthropic Challenges Hackers to Break Its AI

Anthropic unveiled Constitutional Classifiers, claiming to block 95.6% of advanced jailbreak attempts—vastly outperforming standard Claude at 14%. A 3,000-hour bug bounty with zero full breaches led Anthropic to invite public testing. The system trains AI to spot malicious prompts in multiple languages and styles. This underscores a novel approach to AI safety, using AI itself to guard against misuse.

🎉🤝 OpenAI Co-Founder John Schulman Joins Mira Murati’s Startup

John Schulman, a key figure behind ChatGPT, left OpenAI for Mira Murati’s new venture, Thinking Machines Lab. The startup has already attracted several prominent ex-OpenAI researchers. Murati aims to build AI for complex, multi-step tasks, challenging the dominance of incumbents. Schulman’s move signals intensifying competition among next-gen AI players.

🏢🔎 Microsoft’s Suleyman Creates AI Societal Impact Unit

Mustafa Suleyman, Microsoft AI CEO, announced a cross-disciplinary research team bringing in economists, psychologists, and policy experts. The unit will explore AI’s effects on labor, mental health, and more. By integrating social sciences, Microsoft aims to shape ethically grounded AI policies. It’s a sign the tech giant wants to balance innovation with real-world responsibility.

👩‍💻🤖 GitHub Powers Up Copilot with Agentic Features

GitHub introduced new Copilot modes, including an “agent mode” that self-corrects code and suggests terminal commands. Vision allows code generation from images, automatically creating interfaces and alt text. Copilot Edits is now widely available, handling multi-file changes via natural language. Project Padawan, an autonomous coding agent, is set to arrive this year, marking a leap from code suggestions to near-autonomous development.

🇫🇷⚡ Mistral Revamps le Chat Assistant

French AI lab Mistral launched updated iOS and Android apps for “le Chat,” featuring speed enhancements, document processing, code interpretation, and image generation. A new “Flash Answers” feature delivers results 10x faster than competitors. Enterprise options include on-premise setups and custom model integration. Mistral’s focus on rapid responses and flexible deployments addresses key user needs.

📹✨ Pika Labs Launched ‘Pikadditions’ to Add Anything to Any Video

Pika introduced Pikadditions, letting its 12 million users overlay any subject or object onto existing videos. This follows the 2.1 model update, which improved resolution, character control, and camera angles. Creators can now remix favorite movie clips or personal reels by inserting comical or surreal elements. Pika aims to spur unprecedented creativity in social video editing.

🗣️🏠 Amazon Plans Next-Gen Alexa AI This Month

Amazon is revamping Alexa with generative AI powered by Anthropic’s Claude, aiming for smoother, context-aware conversations. Early internal models struggled with latency, prompting Amazon’s pivot to Claude. Over 100 million people actively use Alexa, making reliability vital. The reveal comes just before Apple’s rumored Siri overhaul, intensifying competition in voice assistants.

🍎🎉 Apple Introduces AI-Powered Party Planner

Apple’s new “Invites” app blends AI with existing services like Photos, Music, Maps, and Weather to organise events. Users can generate custom images and text for invitations, then track RSVPs and share photos, even with non-Apple users. It’s Apple’s first standalone AI-driven release, signaling a practical approach to AI-based features. The debut suggests Apple’s gradual but steady expansion in consumer-focused AI.

⚠️🔧 Hugging Face Lays Off 4% of Staff

Hugging Face cut around 10 roles from its sales team, representing about 4% of the company. The move scales back its “Expert Support Program” in favor of more predictable revenue streams like APIs and cloud referrals. Despite a $4.5B valuation and strong community adoption, monetising open-source tech remains a challenge. Hugging Face aims to sharpen its focus on core developer-focused services.

📱🖼️ Snap Unveils High-Speed AI Text-to-Image Model

Snap introduced a mobile-first AI that generates high-resolution images in just 1.4 seconds on an iPhone 16 Pro Max. The technology aims to seamlessly integrate into Snapchat, allowing instant visual creativity. Unlike cloud-heavy counterparts, Snap’s optimisation for mobile hardware sets it apart. The move could herald more on-device AI experiences across social platforms.

🆓🚀 Replit’s AI Agent Now Free to Try

Replit has opened its AI Agent to all users at no cost, including via its mobile app. The platform lets anyone create, develop, and deploy apps without writing code manually. It centralises the entire development cycle—coding, hosting, and collaboration—in one place. This reinforces Replit’s mission to democratise software development through accessible AI tools.

🚀💲 Cursor Code Editor Hits $100M Revenue at Record Pace

Cursor, an AI-driven coding editor, soared from $1M to $100M in revenue faster than any previous app. By leveraging an LLM “wrapper” strategy, it keeps overhead low while attracting many developers. This explosive growth highlights the lucrative potential of layering AI features atop existing large language models. It underscores a broader trend: specialised wrappers can be more profitable than building LLMs from scratch.

🏋️‍♂️🤖 Nvidia’s AI Teaches Robots Athletic Moves

Nvidia and Carnegie Mellon introduced ASAP, letting robots learn motions from simulation, then adapt to real-world physics. Test robots replicated NBA and soccer star moves, reducing motion errors by 53% versus older methods. While hardware remains a bottleneck—motors can overheat—the breakthrough suggests advanced robotics may soon mirror human athletic feats. It’s a leap forward in bridging virtual training and physical execution.

🚕💬 Lyft’s New AI Customer Assistant Powered by Anthropic

Lyft deployed an AI tool using Anthropic’s Claude to handle initial inquiries from riders and drivers. Operating through AWS Bedrock, the system answers common support questions before escalating complex issues to human agents. This mirrors Uber’s earlier partnership with OpenAI for driver-oriented AI features. The deal exemplifies how rideshare giants are turning to generative AI for faster, smarter customer service.

🚫🤖 Texas Governor Bans Deepseek AI on State Devices

Governor Greg Abbott barred Chinese AI model Deepseek from all state-owned devices over data security and intellectual property concerns. The move parallels recent federal-level bans on Deepseek, hinting at broader nationwide restrictions. Critics question whether this intensifies U.S.-China tech tensions. It underscores increasing scrutiny of AI software originating from geopolitical rivals.

💶🌍 EU Invests $56M in Open-Source AI Rival

The European Union is funding OpenEuroLLM, aiming to create an open-source large language model covering all 30 EU languages. Though $56M pales next to Big Tech budgets, the project echoes DeepSeek’s success with lower-cost, open development. It will leverage EU supercomputers and open licensing to spur innovation aligned with European values. This marks the bloc’s pivot from mere AI regulation to direct AI investment.

⚖️🚀 EU Activates First Phase of AI Act

The EU began enforcing Article 5 of its AI Act, banning systems deemed “unacceptably risky,” such as real-time facial recognition in public spaces. The law classifies AI into four tiers, from minimal to outright prohibited. Companies violating the ban face fines up to €35 million. Europe aims to set a global precedent by balancing innovation with strict consumer protections.

🧬⏱️ MIT Researchers Unveil ChromoGen

MIT’s new AI model, ChromoGen, predicts 3D genome structures in minutes instead of days. Rapid mapping of DNA folding offers insights into gene regulation, cell function, and disease mechanisms. This speedup could catalyse breakthroughs in personalised medicine. ChromoGen shows AI’s increasing power to accelerate complex scientific research.

🩺🔬 AI Creates a Database to Help Detect Cancer Earlier

Johns Hopkins researchers unveiled AbdomenAtlas, a 45,000-scan 3D CT dataset annotated for 142 anatomical structures—36 times bigger than any rival. By combining AI and human radiologists, they completed in two years what would have taken 2,500 years manually. The open-source dataset aims to accelerate tumor detection and enhance early diagnosis. It’s a leap toward AI-powered medical breakthroughs.

💊🔎 AI Identifies Life-Saving Drug for Rare Disease

University of Pennsylvania researchers used AI to discover adalimumab’s effectiveness against idiopathic multicentric Castleman’s disease, saving a hospice patient. The machine learning approach analyzed 4,000 potential treatments, honing in on TNF as a key target. Driven by his own struggle with iMCD, lead scientist David Fajgenbaum co-founded Every Cure to advance such discoveries. A clinical trial for another repurposed drug is already in the works.

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