Weekly AI News

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

This week’s headlines are brimming with power plays, bold moves, and global ambitions. The Paris AI summit turned into an unexpected showdown, with the U.S. and UK refusing to sign a pledge while European leaders unleashed massive investment plans. Not to be outdone, Elon Musk tried—and failed—to buy out OpenAI, ruffling plenty of feathers in the process. Meanwhile, Anthropic and Google quietly prepped their newest AI models, focusing on coding efficiency and “reasoning on demand.”

In a more playful corner, AI made a grand entry into Super Bowl commercials. We witnessed everything from black-and-white ChatGPT narratives to Ray-Ban AR glasses that let celebrities record video from a mere flick of the frame. On the creative side, Adobe debuted “commercially safe” video generation, and ByteDance showcased unbelievably real AI-based product demos, pushing the boundaries of marketing.

We’ve also seen legal battles heating up over copyright infringement. Cohere and Ross Intelligence faced lawsuits that could become pivotal references for how AI firms handle data. And as legislative concerns grow around AI chatbots interacting with kids, it’s clear that regulators are ready to draw some lines in the sand.

Happy reading!

⚖️🇫🇷 Tensions Erupt at Paris AI Summit

The Paris summit ended amid disagreements as the U.S. and UK refused to sign a global AI declaration. European leaders countered with big spending pledges, hoping to challenge American dominance. Safety worries took a back seat, replaced by a push for AI growth and deregulation. Critics called it a “missed opportunity” to address critical AI risks.

🏈🤖 AI Joins Super Bowl Prime Time

Super Bowl LIX featured AI-centric commercials from OpenAI, Google, Meta, and more. OpenAI’s first-ever Super Bowl spot likened ChatGPT to historic breakthroughs like electricity and space travel. Google showcased Gemini Live assisting a busy father, while Meta highlighted AI-powered Ray-Ban glasses. The audience response was mixed, but AI stole the advertising spotlight.

🔥🏆 OpenAI Board Rejects Musk’s $97B Takeover Bid

Sam Altman announced the nonprofit board would refuse Elon Musk’s unsolicited $97.4B bid for OpenAI’s assets. Musk’s offer—backed by xAI and big investors—aimed to stop OpenAI’s for-profit shift. Altman called the proposal “embarrassing,” insisting it runs against OpenAI’s mission. Musk’s legal claims over OpenAI’s nonprofit status remain unresolved.

🤝💼 Musk Will Withdraw OpenAI Bid if Profit Plans Halt

Elon Musk says he’ll drop his $97.4B offer for OpenAI if its board abandons converting the nonprofit into a for-profit entity. He argues the current deal undervalues the nonprofit’s assets. OpenAI’s Sam Altman has already labeled Musk’s bid a non-starter. The court battle reflects a deeper fight over who controls the future of advanced AI.

🚀🧠 Sam Altman Talks GPT-5

Speaking at the University of Tokyo, Sam Altman previewed GPT-5, the Stargate Project, and OpenAI’s push into open-source. He claimed OpenAI’s internal model already ranks as the world’s 50th-best programmer—potentially topping the charts by year’s end. Costs are dropping 10x annually, fueling faster AI adoption. Altman believes upcoming models will generate new scientific insights sooner than we think.

🌌💡 GPT-5 Roadmap Unveiled, o3 Scrapped

OpenAI’s CEO “canceled” the planned o3 model in favour of integrating its features into GPT-5. GPT-4.5 (“Orion”) will arrive within weeks as the final “traditional” model before GPT-5’s unified approach. Free users get a standard intelligence tier of GPT-5, while paid tiers unlock advanced capabilities. The streamlined naming aims to reduce confusion and deliver best-fit intelligence automatically.

🧩🔗 ChatGPT o3-mini Changes Its Chain of Thought

OpenAI updated how ChatGPT o3-mini displays its reasoning to increase transparency. However, critics accuse it of copying DeepSeek’s R1 explanation style. OpenAI argues that showing users the model’s thought process helps build trust and fosters better decision-making. The feature underscores ongoing debates about revealing or hiding AI “chain-of-thought” outputs.

🔧🏭 OpenAI Tapes Out First In-House Chip

OpenAI is finalising its first custom AI chip design to reduce reliance on Nvidia. The design will soon head to TSMC for manufacturing, with mass production targeted for 2026. Each initial tape-out can cost tens of millions of dollars—and success on the first try is never guaranteed. A functioning custom chip would mark a major milestone in OpenAI’s hardware strategy.

🏙️🇩🇪 OpenAI Sets Up Shop in Munich

OpenAI is opening a new office in Munich, citing Germany’s leading position in European AI adoption. The country boasts the continent’s largest base of ChatGPT users, paying subscribers, and API developers outside North America. This strategic move strengthens OpenAI’s European presence. Munich will serve as a key hub for regional partnerships and R&D.

 📜⚖️ OpenAI Expands Model Spec, Emphasises Free Expression

OpenAI released a 63-page Model Spec update detailing how ChatGPT should handle controversial topics. A new “chain of command” ensures platform rules override user preferences but also supports adult or sensitive content within safe limits. The Spec aims to reduce AI’s tendency to pander and encourages more candid, truthful responses. It’s shared under CC0 for any AI group to adopt or adapt.

⚡🏆 Gemini Flash 2.0 Tops AI Agent Leaderboard

Galileo Labs launched a tool-usage leaderboard ranking 17 leading LLMs. Google’s Gemini 2.0 Flash took first place with a 0.938 score, beating pricey competitors on speed and complex tasks. Open-source solutions like Mistral’s new Small model also impressed with strong performance at lower costs. As AI agents evolve, specialised benchmarks are becoming vital for comparing real-world capabilities.

📐🥇 DeepMind AI Conquers Math Olympiads

Google DeepMind unveiled AlphaGeometry2, solving 84% of International Mathematical Olympiad geometry problems from the past 25 years—beating the average gold medalist score. By combining a Gemini model with a symbolic engine, it tackled rigorous proofs and intricate geometry challenges. The system generated millions of synthetic theorems for training. This breakthrough points to new AI-driven progress in advanced math and sciences.

📚🔍 Perplexity Launches Deep Research

Perplexity introduced “Deep Research,” which creates in-depth reports on any topic by iteratively searching, reasoning, and coding. It’s free for up to five queries a day for non-subscribers and 500 for Pro users. Deep Research scores 93.9% on SimpleQA and 21.1% on the notoriously tough Humanity’s Last Exam, beating many top models. With near real-time analysis, it’s poised to be a powerful research companion.

⚡🐬 Perplexity Unleashes Ultra-Fast Sonar Model

Perplexity introduced a new version of its Sonar model based on Llama 3.3 (70B), delivering answers at 1,200 tokens per second. Despite the speed boost, Sonar maintains high factual accuracy and outperforms rival systems in user satisfaction. Powered by Cerebras hardware, Sonar aims for near-instant responses. Pro subscribers get immediate access, with API availability coming soon.

🧢🧠 Meta’s Brain-Reading ‘Hat’ for Typing

Meta showcased an external device that detects brain signals to simulate keypresses with about 80% accuracy—no implants required. While still not as fast as a keyboard, it’s a notable step forward in noninvasive brain-computer interfaces. The system paves the way for hands-free text entry. Future iterations could improve speed and reliability for broader use.

🏠🤖 Meta Explores Human-Robot Co-Work in Homes

Meta announced PROJECT PARTNR, a research initiative focused on human-robot collaboration in everyday tasks. Early studies involve chores like cooking and cleaning, aiming to understand how robots and humans can split responsibilities. The project emphasises intuitive cooperation, requiring advanced AI for context and adaptability. It signals Meta’s broader push into practical home robotics.

🏭🇰🇷 Meta Eyes Korean AI Chip Startup Acquisition

Reports suggest Meta is in talks to acquire FuriosaAI, a South Korean chipmaker specialising in AI accelerators. The move could bolster Meta’s custom silicon ambitions, reducing dependence on third-party GPU suppliers. FuriosaAI’s CEO brings experience from Samsung and AMD, indicating strong technical leadership. No official comment yet, but insiders predict a deal soon.

🧠📈 Anthropic Readies Major Claude Upgrade

Anthropic is poised to release a powerful new model that merges traditional language skills with advanced “deep reasoning.” Developers will be able to fine-tune how much computation each query receives, maximising efficiency or detail on demand. The model performs exceptionally on coding tasks and large codebases. Rumors suggest it surpasses OpenAI’s o3 internally but was delayed over safety concerns.

⚖️⚙️ Anthropic’s Hybrid AI for Dynamic Resource Use

Anthropic revealed a sliding-scale approach in its upcoming model, letting developers dial up or down “reasoning power.” Setting the scale to zero transforms it into a standard LLM, while higher levels tackle complex tasks with more tokens and compute. This system offers precise cost control and performance tuning. It contrasts with OpenAI’s simpler three-tier method, sparking debate about optimal resource allocation.

📊🏗️ Anthropic Publishes Economic Index

Anthropic’s new Economic Index reports that AI adoption mostly augments, not automates, human labor. Drawing on data from diverse industries, the study highlights improved productivity without widespread job displacement. It also notes emerging roles in AI oversight and integration. Anthropic aims to provide ongoing insights into the technology’s economic impacts.

💹🚀 Anthropic Projects $34.5B Revenue by 2027

Despite burning $5.6B in cash last year, Anthropic predicts significant revenue growth—up to $3.7B in 2025 and possibly $34.5B by 2027. Analysts say its focus on safer, enterprise-friendly models could attract lucrative partnerships. The ambitious projection pits Anthropic against giants like OpenAI. If realised, it would mark one of the fastest expansions in AI history.

🔎📐 Berkeley’s DeepScaleR Outdoes OpenAI’s o1

UC Berkeley researchers debuted DeepScaleR, a tiny 1.5B-parameter open-source model outperforming OpenAI’s o1-Preview in math reasoning. Despite its smaller size, DeepScaleR showcases efficient architecture and specialised training. Its success fuels optimism about more accessible, high-performance models. The project aims to advance cost-effective AI research outside big tech.

🎞️🛡️ Adobe Rolls Out IP-Safe AI Video Generator

Adobe launched the Firefly Video Model, touting it as the first commercially safe AI video generator. Trained only on licensed and public-domain content, the tool assures legal clarity for business use. Users can produce 1080p clips, control shots, and sync audio with minimal risk of IP violations. Two new subscription tiers ($9.99 and $29.99 monthly) offer flexible creation plans.

🏮🎨 ByteDance & HKU Debut Goku, Goku+ Visual Models

ByteDance and the University of Hong Kong unveiled Goku and Goku+, bridging image and video generation with advanced “rectified flow” for smooth transitions. Goku+ focuses on advertising and creating photorealistic product demos and avatars. Both models set new quality records in major benchmarks. Their commercial orientation demonstrates the rapid evolution of AI-generated marketing content. https://saiyan-world.github.io/goku/

🌅🎥 Luma AI Reveals Image-to-Video with Ray2

Luma AI showcased an update to its Ray2 model, now capable of producing highly realistic video from static images. The new system analyses depth and motion cues to generate fluid camera movement. Early demos show near-cinematic results, fueling excitement for creative professionals. It pushes the boundaries of image-based 3D generation for both casual and commercial use.

💡🤖 Apple’s Expressive Lamp Bot for Home AI

Apple researchers introduced ELEGNT, a concept that mixes basic robotic functions with emotive gestures using a lamp-like device reminiscent of Pixar’s Luxo Jr. By nodding, looking around, or “shrugging,” the bot communicates intentions in a more human-friendly way. Siri integration enables voice interaction. Initial tests suggest expressive movements build trust and comfort with in-home robots.

🍎🏯 Apple Partners with Alibaba for China AI Features

Apple teamed up with Alibaba to develop AI functionalities specifically for Chinese iPhone users. The goal is to stay competitive amid slowing iPhone sales and rising domestic rivals like Huawei. Both companies submitted their joint AI features for regulatory approval in China. Apple’s local collaboration underscores the region’s critical importance for its global strategy.

🤝🇨🇳 Alibaba Chairman Confirms Apple Alliance

Alibaba’s Joe Tsai affirmed that Apple chose his company to power AI features for Chinese iPhones. Calling it an “honor,” Tsai noted Apple’s careful selection process and compliance with China’s regulatory requirements. He also hinted that localised models could become the norm, as each region imposes unique rules. The news boosted Alibaba’s stock over multiple trading sessions.

🎬⚡ YouTube Integrates Veo 2 AI Into Shorts

YouTube is rolling out Google DeepMind’s Veo 2 model for text-to-video in Shorts. Users can now create custom clips and backgrounds simply by typing prompts. The update accelerates the Dream Screen feature with more realistic scene physics and Google’s SynthID watermarking for transparency. It’s initially available in the U.S., Canada, Australia, and New Zealand.

🏆🌐 YouTube Empowers Creators with AI Upgrades

YouTube CEO Neal Mohan announced expanded AI tools: automated dubbing for all monetised creators, better AI-based content detection, and improved Dream Screen integration. The new system can identify and manage unauthorised AI-generated copies of creators’ work. Auto-dubbed videos gained 40% extra watch time from foreign-language audiences. It’s another big step in YouTube’s AI-first evolution.

🚗📈 BYD’s Stock Soars on New ADAS with DeepSeek

Chinese EV maker BYD saw its Hong Kong share price jump 4% after revealing a driver-assistance technology co-developed with DeepSeek. The system promises advanced object detection, lane guidance, and real-time route adaptation. Analysts say the partnership could give BYD a competitive edge in the booming EV market. The milestone underscores China’s drive to integrate cutting-edge AI in automotive solutions.

⚙️🤖 Cisco Unveils AI Renewals Agent with Mistral

Cisco and Mistral AI co-developed an agent to streamline customer contract renewals. It pulls data from 50+ sources, analyses client sentiment, and automates relevant workflows. Cisco estimates it can cut proposal prep time by 20% or more. The tool showcases how generative AI can handle specialised business processes, hinting at broader enterprise applications.

📊💡 PwC Finds 79% of Its US Staff Use GenAI

PwC revealed that over three-quarters of its 75,000 U.S. employees regularly use generative AI at work. On average, 26% consult these AI tools at least once a day, with many relying on them for research, drafting, or data analysis. PwC’s findings underscore the rapid adoption of AI in professional services, reshaping daily workflows.

🟠🤝 Orange & Mistral AI Boost European AI Development

Orange and Mistral AI announced a strategic partnership to advance AI research in Europe. Their collaboration targets network optimisation and integrating Mistral’s models into Orange’s products for enterprise clients. Both companies see the deal as a major step toward a stronger European AI ecosystem. It aligns with EU calls for homegrown AI champions to compete globally.

👾🛡️ Alphabet, OpenAI, Roblox, and Discord Launch ROOST

Tech giants unveiled ROOST, a $27M open-source initiative to fight online child exploitation. The project aims to develop free AI tools to detect harmful content, track suspicious activity, and create safer digital experiences. By sharing methodologies and code, they hope to help smaller platforms implement robust safety measures. ROOST signals growing industry cooperation on child-protection tech.

💸⚙️ Big Tech to Spend $325B on AI in 2025

Meta, Microsoft, Amazon, and Alphabet collectively plan to invest $325 billion in capital expenditures next year, largely for AI infrastructure. The figure marks a 46% jump from the previous year. Critics warn of ballooning bills with uncertain returns, but the companies see it as essential to maintain competitive AI capabilities. This arms race could shape data center and chip innovation for years.

🇫🇷💶 Macron Announces €109B French AI Investment

During the Paris AI summit, President Macron revealed €109 billion in private-sector commitments to bolster France’s AI sector. Major chunks come from Canadian firm Brookfield and the UAE, partly earmarked for a 1-gigawatt data center. The plan aims to position France as a top AI hub in Europe, bridging research, infrastructure, and industrial applications.

🏛️👦 California Bill Would Tell Kids “Chatbots Aren’t Humans”

State lawmakers introduced SB 243, requiring AI platforms to remind minors that chatbots are artificial—not real people. The bill seeks to protect children from AI’s “addictive” and “influential” effects and mandates annual reporting on detected suicidal ideation. Supporters say it’s a necessary guardrail, but critics argue it could stifle innovation or overburden developers.

⚖️📚 Thomson Reuters Secures Major AI Copyright Win

In a landmark case, a judge ruled that Ross Intelligence infringed Thomson Reuters’ copyrights by using Westlaw content to train an AI without permission. This decision sets an early precedent on whether AI scraping qualifies as “fair use.” With similar suits pending against OpenAI and Microsoft, legal watchers see it as a pivotal moment for AI training data practices.

📰⚠️ Publishers Sue Cohere for Copyright Infringement

Condé Nast, The Atlantic, and The Guardian filed suit against AI startup Cohere, claiming it uses copyrighted articles—even paywalled ones—for training. Cohere denies wrongdoing, saying it follows best practices to avoid IP violations. This is the latest in a series of publisher-led lawsuits against AI firms over data sourcing. The outcome could reshape how AI companies gather and process text.

🚨🎙️ Sophisticated AI Scam Targets Italy

A high-tech scam in Italy cloned the voice of the country’s Defense Minister to trick wealthy individuals into paying ransom for allegedly imprisoned journalists. At least one business leader reportedly transferred €1 million. The incident underscores the rising threat of deepfake voice technology. Authorities warn companies to implement stronger verification processes.

✍️🚫 Thousands Urge Christie’s to Cancel AI Art Auction

Nearly 4,000 signatories penned an open letter objecting to Christie’s AI-focused art sale. They argue that some featured artworks were trained on copyrighted images without permission, exploiting human creators. Christie’s expects the auction to exceed $600,000 in sales. The debate spotlights ongoing tensions over ethical AI usage in the art world.

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