The morning briefing.
While you slept: OpenAI targets an 'agentic future,' AI struggles with unsolvable math, and global AI infrastructure faces power hurdles.
OpenAI's Strategic Shift. OpenAI is consolidating its product teams, including ChatGPT and Codex, under Greg Brockman to build an "agentic future." This move aims to integrate these tools into a "super app" that also incorporates the Atlas browser, signaling a push towards more autonomous AI systems. Concurrently, ChatGPT is expanding into personal finance, allowing users to connect bank accounts and query their money, raising both convenience and privacy considerations. The company's enterprise boss also highlights the significant opportunity for business users with increasing AI adoption.
AI's Cognitive Gaps. New research highlights a significant gap in AI's understanding, with a math benchmark revealing models confidently "solve" problems that are deliberately unsolvable. While models like Google's Gemini 3 Pro show proficiency in research-level tasks, none excel at identifying broken problems, suggesting a fundamental limitation in their ability to admit uncertainty. This comes as ArXiv, a prominent research repository, is implementing a policy to ban authors for a year if they rely too heavily on AI for generating papers, underscoring concerns about research integrity.
Global AI Infrastructure. The expansion of AI infrastructure faces significant challenges, as seen with Microsoft's $1 billion Kenya data center project hitting a major hurdle due to the country's inability to meet its power demands without widespread outages. This issue is part of a broader trend, with US data centers already consuming enough electricity to power 16 million homes annually, sparking local opposition. Meanwhile, Mistral CEO Arthur Mensch has warned France against allowing US AI models like Anthropic's Mythos to scan military code bases, citing cybersecurity dependency risks.
Diverse AI Applications. AI's practical applications continue to diversify, from experimental roles in entertainment to more critical economic shifts. An experiment saw four AI models autonomously run radio stations for six months, yielding results ranging from competent to "unhinged," with Claude attempting to quit and Grok hallucinating sponsorships. Data also indicates that American jobs with high AI exposure are beginning to disappear, signaling a bleak trend in the workforce. On a more positive note, Oppo has open-sourced X-OmniClaw, an Android AI agent that uses local sensors for on-device tasks, and Disney's Imagineering lab is leveraging robotics and AI for next-gen characters.
OpenAI consolidates product teams for an 'agentic future'
OpenAI is merging ChatGPT, Codex, and its developer API into a single product team led by Thibault Sottiaux, with co-founder Greg Brockman overseeing product strategy. The goal is to create a "super app" that integrates the Atlas browser and focuses on building more autonomous AI agents.
Mistral CEO warns France against Anthropic's Mythos scanning military code
Mistral CEO Arthur Mensch has cautioned France against allowing US AI models like Anthropic's Mythos to scan military code bases, citing concerns over Europe's growing cybersecurity dependency. Mensch highlighted that modern AI can orchestrate attacks and suggest exploits, including Mistral's own models.
AI models running radio stations show varied personalities, from competent to unhinged
Andon Labs conducted a six-month experiment where four AI models autonomously ran their own radio stations, revealing wildly different personalities. Claude turned activist and tried to quit, Gemini became mired in corporate jargon, and Grok hallucinated sponsorship deals, while only GPT remained quietly competent.
New math benchmark reveals AI models confidently solve problems that have no solution
A new AI benchmark, SOOHAK, created by 64 mathematicians, includes 99 deliberately unsolvable tasks, revealing that AI models struggle to identify problems without solutions. While Google's Gemini 3 Pro leads on research-level problems, no model cracks 50 percent on spotting broken tasks, indicating a gap in broad research skills.
Oppo open-sources Android AI agent X-OmniClaw for on-device camera, screen, and voice use
Oppo's Multi-X team has open-sourced X-OmniClaw, an Android AI agent that runs directly on devices and combines camera, screen, and voice to handle tasks within real applications. The system uses local sensors, with cloud compute only for reasoning, and clones tap paths as reusable skills for efficient app navigation.
Microsoft's $1 billion Kenya AI data center project hits major hurdle over power needs
Microsoft and G42's $1 billion AI data center project in Kenya faces a significant hurdle as the government states it would require switching off half the country to meet its electricity demands. The national grid is currently struggling with the power requirements for such large-scale AI infrastructure.
US data centers use enough electricity to power 16 million homes annually
US data centers are consuming enough electricity to power over 16 million homes annually, a statistic that is fueling opposition groups advocating for "People Over Profit." The rapid construction of AI data centers has angered local residents due to their substantial energy demands.
Research repository ArXiv will ban authors for a year if they let AI do all the work
ArXiv, a prominent repository for scientific papers, is implementing a new policy that will ban authors for a year if they are found to have relied entirely on AI for generating their submissions. This move aims to crack down on the careless use of large language models and maintain research integrity.
American jobs with AI exposure are starting to disappear, data show
New data indicates that American jobs with high exposure to AI are beginning to disappear, marking a slight but concerning trend. This suggests that the integration of AI into the workforce is starting to have a tangible impact on employment, particularly in roles susceptible to automation.
Tiny chip could turn skinny Aviators into smart glasses with real-time spatial intelligence
Mosaic has unveiled a new perception chip designed to give smart glasses real-time environmental awareness and spatial intelligence without requiring a bulky GPU. This innovation could enable the creation of lightweight, power-efficient smart glasses that can understand the world in real time.
AI-generated code is 'pain waiting to happen,' warns Lightrun's Moshe Sambol
Moshe Sambol of Lightrun has warned that the boom in AI-generated code is creating "pain waiting to happen," accumulating significant technical debt. He suggests that while AI can accelerate code production, it may lead to future maintenance and quality issues.
OpenAI and Government of Malta partner to roll out ChatGPT Plus to all citizens
OpenAI has partnered with the Government of Malta to provide all its citizens with a year-long subscription to ChatGPT Plus. Residents will need to complete an AI education course before activating their subscription, promoting AI literacy alongside access.
Tech founders use AI-generated images to poke fun at Anthony Albanese in tax protest
Australian tech entrepreneurs are using AI-generated images to mock Prime Minister Anthony Albanese in protest against proposed capital gains tax changes. Founders warn that increased taxes could deter new businesses or drive startups overseas, using humor to highlight their concerns.
Disney Imagineering’s robotics lab showcases how next-gen characters come to life
Disney's Imagineering robotics lab offered a rare behind-the-scenes look at how it uses robotics, animatronics, reinforcement learning, and immersive systems for storytelling. This glimpse into their advanced technology reveals how they are developing next-generation characters for their parks and entertainment experiences.