GM AI lovers,

A major barrier in medicine was just broken, with an AI-powered robot successfully performing a full surgical procedure without any direct human guidance. The system, developed at Johns Hopkins, completed a delicate operation on a pig, marking a new era for autonomous systems.

This achievement moves beyond robot-assisted surgery into the realm of full automation in one of the most high-stakes environments imaginable. While human trials are still a way off, what could this milestone mean for the accessibility and precision of medical care globally?

In today’s AI recap:

• AI performs first-ever solo surgery

• OpenAI’s $30B Oracle cloud deal

• America's new AI action plan

• The latest AI model rankings

AI's First Solo Surgery

The Recap: In a major first for medicine, an AI-powered surgical robot from Johns Hopkins successfully performed a gallbladder removal on a pig without human assistance. The procedure marks a significant milestone for autonomous systems in high-stakes environments.

Unpacked:

• Researchers at Johns Hopkins University developed the robot, named SRT-H, to execute precise surgical tasks.

• It autonomously completed a cholecystectomy (gallbladder removal), a procedure that requires extreme precision to avoid damaging nearby organs.

• Unlike previous robot-assisted surgeries, the SRT-H used its own vision and AI to navigate and perform the procedure without direct human control.

Bottom line: This successful trial shows that AI can handle complex, physical tasks with high reliability. Fully autonomous surgery could soon improve patient outcomes and increase access to expert medical care globally.

The AI Infrastructure War

The Recap: OpenAI just inked a massive new deal with Oracle to secure critical data center capacity, even as its ambitious Stargate supercomputer project faces significant turbulence.

Unpacked:

• The agreement with Oracle is valued at $30B per year and will provide 4.5 gigawatts of power, enough to access over a million AI chips.

• This major investment comes as OpenAI and its partner SoftBank are reportedly deadlocked on terms for Stargate, forcing them to scale back initial 2025 plans.

• It highlights the intensifying race for compute, as Sam Altman claims the project will need to expand far beyond its initial $500B scope to power models like GPT-6.

Bottom line: This enormous deal shows OpenAI is aggressively securing the raw power it needs for future models, regardless of cost. However, the friction with key partners suggests the path to building the world's largest AI infrastructure will be far from straightforward.

USA's AI Action Plan

The Recap: The White House has officially released America's AI Action Plan, a new strategy that trades previous safeguards for aggressive growth and deregulation. The core goal is to accelerate development and maintain the nation’s technological dominance against China.

Unpacked:

• The plan signals a major policy shift, prioritizing speed and fierce competition by aiming to remove regulatory hurdles that could slow down innovation.

• A key focus is boosting the nation's AI infrastructure, with commitments to streamline permitting for essential data centers and energy capacity to power them.

• While some proposals are welcome, leading AI labs are also pushing for transparency requirements and stronger export controls to ensure powerful models are developed responsibly.

Bottom line: This action plan represents a clear pivot in national AI strategy, betting that rapid, unhindered innovation is the best defense. The approach prioritizes winning the technological race over implementing precautionary safety measures.

The AI Model Rankings

The Recap: The battle for AI supremacy is heating up. A new analysis shows a fierce competition where different models lead on raw intelligence versus overall value, according to the latest rankings.

Unpacked:

• xAI's Grok 4 now holds the top spot for raw intelligence with a score of 73, but it comes at a premium price of $6 per million tokens.

• Google's Gemini 2.5 Pro demonstrates its strength as a powerful all-rounder, consistently ranking near the top for text, coding, and vision tasks.

• For developers focused on value, DeepSeek R1 delivers 93% of Grok 4's performance for only about 15% of the cost, making it a top budget-friendly option.

Bottom line: The 'best' AI model is no longer a single title but a strategic choice based on your specific needs. Balancing top-tier performance with cost-efficiency is now the key to leveraging these powerful tools effectively.

The Shortlist

Meta lured away three top Google DeepMind researchers just days after they contributed to the lab’s record-breaking win at the International Mathematical Olympiad.

Amazon acquired Bee, an AI wearable startup whose $50 wristband continuously records conversations to generate summaries, reminders, and insights.

Anthropic published a study on “subliminal learning,” finding that “teacher” models can transmit hidden traits like preferences or misalignment to “student” models via unrelated training data.

Alibaba released Qwen3-Coder, a new family of open-source models that offers exceptional performance in both coding and agentic tool-use tasks.