Personal Superintelligence
In a new blog post, Mark Zuckerberg outlined his belief that superintelligence is “within sight” and that Meta will work towards building a "personal superintelligence" for everyone.
Between the lines:
Meta is framing the AI race as a philosophical battle between individual empowerment and centralized control. Zuckerberg’s vision is focused on personal AI assistants that know you deeply and help you achieve your goals.
Of course, this is all occurring against the backdrop of Meta spending billions on its Superintelligence Lab, ruthlessly poaching top researchers from other organizations, and retreating from previous open source commitments.
However, some inside Meta (and certainly many outside) are still uncertain about the lab's strategy and even how it will impact existing AI teams such as FAIR (fundamental AI research), which already employ scores of AI researchers.
Elsewhere in the FAANG free-for-all:
Google rolls out new AI Mode features including Search Live with video input, Canvas for planning, and image/PDF uploading capabilities.
Microsoft Edge adds Copilot Mode, an experimental feature that creates an agentic AI browsing experience by anticipating user actions.
Meta is piloting letting job candidates use AI assistants during coding interviews and seeking employee volunteers for mock AI-enabled interviews.
Microsoft previews Copilot Appearance, a virtual character with real-time expressions and conversational memory.
And Google is testing Opal, a vibe-coding tool that lets users create mini web apps using text prompts or remix existing apps.
Elsewhere in the AI war for talent:
The AI talent wars have reached the New York Times, with new details around Meta offering potential hires 30,000 GPUs toward their AI research priorities.
Apple lost its fourth AI researcher to Meta's superintelligence team in a month as the AFM team remains in flux despite marginal pay increases.
Meta approached over a dozen people at Thinking Machines Lab with offers including $1B+ over multiple years, but not a single person has left for Meta so far.
And Mark Zuckerberg is considering more shake-ups to Meta's AI strategy after losing faith in the team partly due to a controversy over whether it gamed AI benchmarks.
Unicorn Autopsy
Builder.ai, once valued at $1.5 billion, promised to use AI to revolutionize app development - but after its founder was ousted for fraudulently inflating revenue, the company filed for bankruptcy last month. Now, we have more insight into what went wrong with the first major gen AI casualty.
Why it matters:
While Builder.ai used AI for basic tasks, most of the development work was outsourced to humans in countries like India and Ukraine.
It's textbook "AI washing" - where startups falsely promote products as AI-driven to attract funding and attention. While not a new phenomenon - it certainly predates ChatGPT - it's gotten bad enough to warrant crackdowns from the SEC and FTC.
This is unlikely to be the last big AI company with nothing behind the curtain - with how hot the AI sector is, even major investors like Microsoft can get burned. Some have even argued there was a severe lack of due diligence in this case (specifically with founder Sachin Dev Duggal).
Elsewhere in AI anxiety:
Repeated warnings of AI-driven layoffs fuel anxiety among workers, who report feeling pressured to accept pay cuts or worse conditions to keep their jobs.
OpenAI's ChatGPT Agent casually clicks through "I am not a robot" verification tests.
DOGE has built an AI tool to cut 50% of federal regulations, or around 100,000 rules, by the first anniversary of Trump's inauguration.
The video game industry is embracing AI-enhanced gameplay while major studios cut employees, worrying some developers despite the industry's use of AI since the 1980s.
And at the International Mathematical Olympiad, 26 students got higher scores than DeepMind and OpenAI models - possibly the last time humans beat AI at the exam.
World AI Conference
China unveiled its global AI vision at the World AI Conference in Shanghai, positioning itself as a champion of open-source development and international cooperation in direct contrast to Trump's "America First" AI strategy.
The big picture:
By framing itself as the champion of the Global South and "open innovation," China is attempting to organize against US export controls by positioning American restrictions as anti-competitive.
And while U.S. chip restrictions have slowed China's progress, they've also accelerated a "whole-nation effort" involving massive state spending, to catch up to existing infrastructure.
For example, China's manufacturing muscle and massive domestic market allow for rapid testing and deployment at scale, particularly in robotics. But concerns remain about over-investment in sectors without clear commercialization paths.
For now, the strategy appears to be working - Chinese companies showcased products built on domestic open-source models from Alibaba and DeepSeek, marking a shift away from reliance on Meta's Llama or OpenAI's GPT.
Elsewhere in AI geopolitics:
The Trump administration is reportedly reconsidering replacing Biden-era AI export rules on chips to ease its tough-on-China approach.
xAI will sign the EU AI Code's safety chapter, but says other requirements hurt innovation, and copyright terms are overreach.
China's Cyberspace Administration questioned Nvidia over H20 chip security concerns after US lawmakers urged adding tracking features to exported AI chips.
And Italy opened an investigation into Meta for potentially violating EU competition rules by integrating Meta AI into WhatsApp without user consent.
Elsewhere in frontier models:
Google DeepMind unveiled AlphaEarth Foundations, an AI model that unifies petabytes of Earth observation data to help scientists create detailed maps on demand.
Harmonic, an AI math startup co-founded by Robinhood CEO Vlad Tenev, unveiled its model Aristotle, supposedly achieving gold medal performance on the 2025 IMO.
Z.ai, formerly known as Zhipu and backed by Tencent, released GLM-4.5, an open-source AI model that it claims is cheaper than DeepSeek.
Alibaba released its Qwen3-235B-A22B-Thinking-2507 reasoning LLM on Hugging Face, topping several benchmarks as the company moves away from hybrid reasoning models.
And sources report that GPT-5 improves coding performance, particularly in practical software engineering tasks, outperforming prior OpenAI models and Claude Sonnet 4.
Things happen
AI has transformed data center design amid ongoing building frenzy. OneChronos wants to create smart markets for trading GPU compute like oil. Runway and Luma AI expect robotics revenue to eclipse studios. An interview with Anthropic CEO Dario Amodei on how his father's death shaped his path. Anthropic debuts new rate limits for Claude Pro power users. Imax partners with Runway to screen AI films at 10 US locations. Anysphere launches Bugbot, an AI debugging tool for $40/month. AI boom reenergizes San Francisco. Huawei outperforms Nvidia on some metrics. Startups like Positron and Groq develop inference chips to rival Nvidia. China's Unitree markets a humanoid robot for $5,900. Grads earn $30-$160/hour training AI. Auterion to ship 33K AI drone kits to Ukraine. Premier League adds AI fan tools. Uber Eats adds AI menu tools and pays users for food photos. Amazon-backed Showrunner launches AI streaming service for $10-$40/month. Google signs EU's AI Code of Practice. Yelp creates AI restaurant videos. Adobe adds "Harmonize" compositing to Photoshop. Google brings AI Mode to UK. Microsoft plans record $30B+ quarterly capex for AI infrastructure. Nadella's layoff memo hints at AI's workforce impact. Alibaba previews AI glasses for 2025. Jeff Dean backs 37 AI startups as angel investor. AI agents and apps threaten Google's ad business by reducing clicks. Enough AI copilots, we need AI HUDs. Crush: Glamorous AI coding agent for your terminal. AI is a floor raiser, not a ceiling raiser. Six principles for production AI agents. Irrelevant cat facts increase LLM math errors by 300%. Graduate job listings plunge since 2022. AI bunnies on trampolines cause TikTok crisis of confidence. ChatGPT helps neurodivergent people navigate social encounters. Chinese campuses show greater AI enthusiasm than US and UK counterparts. People create AI vision boards for goals. Profiles of OpenAI's research heads Mark Chen and Jakub Pachocki. OpenAI launches Stargate Norway, its first European Stargate initiative. Trump administration gets 60+ companies to pledge seamless health data sharing. OpenAI debuts ChatGPT study mode. Microsoft seeks ongoing OpenAI access post-AGI. Johns Hopkins Press licenses books for AI training citing closing window. Amazon pays NYT $20-25M annually for AI content. Developer AI trust drops to 33% despite 84% usage. Only 37% use AI for work tasks.
