PLUS: MIT's fix for 'untrainable' networks, the AI bubble, and Altman's space job future
Good morning
A major breakthrough in robotics has arrived, and it's smaller than a grain of salt. Researchers have successfully developed the first fully autonomous, programmable microrobots, capable of navigating and making decisions entirely on their own.
Because these microscopic agents can be mass-produced for pennies, the potential for deploying them in vast swarms is now a reality. What new frontiers could be explored when intelligent machines can operate inside the human body or monitor our environment at an unprecedented scale?
In today’s Next in AI:
The world's first salt-grain-sized autonomous robots
MIT’s new fix for ‘untrainable’ neural networks
The trillion-dollar AI investment bubble
Sam Altman on AI's role in future space jobs
The Incredible Shrinking Robots
Next in AI: Researchers have developed the world's first fully autonomous, programmable microrobots, each smaller than a grain of salt. This breakthrough, detailed in Science Robotics, enables tiny machines to sense their environment, make decisions, and navigate through liquids on their own.
Decoded:
Instead of tiny limbs, the robots generate an electric field to move fluid, creating a propulsion system with no moving parts that overcomes the challenges of microscale physics.
Each robot runs on a microscopic onboard computer that consumes over 100,000 times less power than a smartwatch, a necessary feat to operate on just 75 nanowatts from its tiny solar panels.
Built using standard computer chip fabrication, these robots can be mass-produced for as little as a penny per unit, paving the way for deploying large swarms for complex tasks.
Why It Matters: This work provides a foundation for creating intelligent microscopic agents that can operate in previously inaccessible environments. Future applications could range from exploring internal biological systems to performing environmental monitoring at an unprecedented scale.
MIT Teaches the 'Untrainable'
Next in AI: New findings from MIT's CSAIL show that previously "untrainable" neural networks can learn effectively when guided by another network. The method transfers structural knowledge, unlocking the potential of previously dismissed architectures.
Decoded:
Unlike knowledge distillation, which copies outputs, guidance works by encouraging a target network to match the internal representations of a guide network.
The effect is so powerful that even a short practice session before main training, using random noise as input, helps networks avoid overfitting and improve performance.
Even untrained guide networks are effective, showing they already contain valuable architectural biases that can be transferred to help other networks learn.
Why It Matters: This research suggests a network's failure may stem from a poor starting point rather than a flawed design. It opens the door to using a wider variety of AI architectures, potentially leading to more efficient and specialized models.
The Trillion-Dollar AI Bubble
Next in AI: The AI industry is showing classic signs of a financial bubble, with trillions of dollars in investment fueling intense hype. According to analysis from The Guardian, an eventual market correction could create a critical opportunity to implement more thoughtful regulation.
Decoded:
The sheer scale of investment is staggering, with OpenAI CEO Sam Altman spearheading commitments valued at around $1.5 trillion to build out AI infrastructure.
Some tech leaders view this as a “good” kind of bubble that finances critical infrastructure and expands knowledge, benefits that could outlast the inevitable financial fallout.
This financial frenzy is also fueling a geopolitical race, with the U.S. aiming for a major leap in general AI while China focuses on wider, faster implementation of current AI across its economy.
Why It Matters: The current investment frenzy echoes past tech booms and is likely unsustainable. A market correction seems inevitable, but it could clear the way for building AI that truly serves human interests, not just market hype.
Altman's Next Frontier: Space
Next in AI: OpenAI CEO Sam Altman predicts that within a decade, AI won't just take jobs—it will create entirely new, high-paying roles for college graduates to explore the solar system.
Decoded:
Altman's forecast provides an optimistic counterpoint to the widespread anxiety that AI will eliminate jobs, suggesting it will instead create roles we can barely imagine today.
While it sounds like science fiction, related fields are already booming; aerospace engineer jobs are growing faster than the national average with salaries already exceeding $130,000 annually.
This vision extends beyond space, as Altman also believes powerful AI tools will soon enable individuals to launch one-person billion-dollar companies from scratch.
Why It Matters: Altman's vision frames AI as a catalyst for creating entirely new industries, not just disrupting existing ones. The biggest opportunities may lie in careers and roles that don't even exist yet.
AI Pulse
Google discounted its annual 2 TB and AI Pro plans by 50% for new subscribers, offering access to Gemini 3 Pro and expanded generative AI features for 2026.
Gemini achieved a gold-medal-level performance at the International Collegiate Programming Contest World Finals, demonstrating state-of-the-art reasoning and problem-solving capabilities.
Google collaborated with other leading AI labs to form the Agentic AI Foundation, an initiative aimed at creating open standards for a responsible and interoperable future for agentic AI systems.