AI Daily Briefing — 2026-05-25

At Google I/O, DeepMind CEO Demis Hassabis declared humanity stands at the “foothills of the singularity,” a moment when AI surpasses human intelligence. Yet his speech focused on WeatherNext, a tool that predicted Hurricane Melissa’s landfall, potentially saving lives. This contrast between grand vision and practical application reveals a growing tension in AI-driven science: specialized tools versus agentic systems that could one day conduct research autonomously.

Agentic AI, powered by large language models, is gaining traction. Google Cloud’s Pushmeet Kohli recently wrote that AI is moving from facilitating science to “beginning to do science.” This shift challenges the value of hyper-specialized tools like AlphaFold, which won a Nobel Prize, or WeatherNext. If AI can independently drive discoveries, investing in narrow applications may seem less justified.

Despite this, Google hasn’t abandoned specialized tools. AlphaGenome and AlphaEarth Foundations launched last summer, and AlphaFold remains widely used by over three million researchers. However, signs of realignment exist: Nobel laureate John Jumper now works on AI coding, not science tools, as Google competes with rivals like OpenAI. This prioritization suggests agentic science is becoming the focus, with coding skills essential for autonomous systems.

OpenAI recently announced an agentic researcher system showing promise, and the industry is betting on AI that collaborates with humans as peers—or surpasses them. While specialized tools still save lives, the path forward increasingly favors AI that can think and act independently, reshaping the future of scientific discovery.

Elon Musk appears to be moving away from his long-standing vision of a solar-powered economy on Earth, according to a recent SpaceX IPO filing. The document reveals that Musk’s AI company, xAI, is powering its data centers with unregulated natural gas turbines and plans to spend $2.8 billion more on fossil fuel infrastructure. This marks a stark departure from Tesla’s original Master Plan, which aimed to transition the world from hydrocarbons to a solar electric economy. While xAI has purchased $697 million in Tesla Megapacks for peak load management, it has not invested significantly in Tesla’s solar panels for terrestrial use.

Instead, the filing emphasizes space-based solar power as the future for AI data centers. SpaceX claims orbital solar arrays can generate over five times the energy of ground-based panels due to constant sunlight. Musk and other Silicon Valley leaders have grown interested in launching server racks into orbit, bypassing terrestrial opposition and power constraints. However, the economics remain daunting: power costs for satellites are far higher than for Earth-based centers, and protecting hardware in space is expensive and complex.

SpaceX argues that AI compute demand could reach terawatt-scale annually, far outstripping current data center capacity of about 40 gigawatts. Musk’s first-principles reasoning suggests he views xAI’s current fossil-fueled data centers as temporary, expecting SpaceX to soon deploy gigawatts of orbital servers. The risk, critics note, is that this space-based solution may not materialize as quickly or cheaply as hoped, leaving Musk’s clean energy legacy in question.

This pivot underscores a broader tension in Musk’s empire: while he built his reputation on sustainable energy, his AI ambitions now lean heavily on fossil fuels and speculative space technology.

At a recent developer event in London hosted by Anthropic, nearly half the attendees admitted they had shipped code written entirely by AI—often without even reading it first. The tool, called Code with Claude, represents a growing trend where developers increasingly hand over their work to artificial intelligence. Anthropic says it wants to push automation to its limits, but critics warn that blindly trusting AI-generated code could lead to serious problems, including security flaws and unreliable software.

Meanwhile, a controversial new sporting event called the Enhanced Games is set to debut in Las Vegas, allowing athletes to use performance-enhancing drugs. Organizers say the goal is to push human performance to new extremes, reflecting a broader cultural obsession with optimization and longevity. The event fits into a 2026 zeitgeist where consumers are encouraged to enhance everything from their bodies to their babies.

At Google I/O, DeepMind CEO Demis Hassabis declared that humanity stands “in the foothills of the singularity.” The company unveiled Gemini for Science, an AI system designed to eventually conduct cutting-edge research autonomously. This marks a shift away from specialized models like WeatherNext toward more general, agent-driven systems. Meanwhile, researchers at Google DeepMind, World Labs, and Meta are pushing forward with “world models”—AI systems that aim to understand physical reality in a more human-like way.

Google’s recent overhaul of its search engine, which prioritizes AI-generated summaries over traditional links, has hit an unexpected snag. Users searching for the word “disregard” are now met with a large blank space and a single AI reply that offers no useful information. The Merriam-Webster definition is still present, but buried far down the page, forcing users to scroll past the empty block.

Social media users have criticized the change, pointing out that the AI response serves no purpose for such a straightforward query. In contrast, Bing’s less aggressive AI summaries provide a more helpful result for the same search. One tech journalist noted that this marks the first time in nearly 15 years that a Bing search has proven more valuable than Google’s.

The incident highlights the challenges of rolling out AI features at Google’s massive scale. While the company aims to enhance search with AI, edge cases like “disregard” reveal that the new system can sometimes break basic functionality, leaving users with a frustrating experience.

This Sunday, 42 athletes will compete in Las Vegas at the inaugural Enhanced Games, a controversial event where performance-enhancing drugs are not only allowed but encouraged. Organizers claim all substances are FDA-approved and medically supervised, with a $25 million prize pool incentivizing world records. Critics, however, warn that FDA approval does not guarantee safety for healthy athletes, citing risks like liver damage and diabetes from steroids and growth hormones.

The games feature swimming, track and field, weightlifting, and strongman events. Many competitors are former Olympic medalists or national record holders, drawn by salaries and the chance to openly experiment with banned substances. The event also embraces “technological doping,” such as polyurethane swimsuits banned by the Olympics since 2009, which helped swimmer Kristian Gkolomeev break a record in a recent trial.

This spectacle reflects a broader cultural shift in 2026, where optimization—from longevity drugs to extreme fitness—dominates consumer trends. The Enhanced Games fit neatly into this ethos, challenging traditional notions of fairness and safety. As the world watches, the question remains: will these games redefine human potential or expose the dangers of unchecked enhancement?

Source: MIT Technology Review AI