Dear Readers,

What do artificial blood from Japan, self-improving AI systems and new standards in medical model evaluation have in common? They all show that we are on the threshold of a new era - one in which technological systems are no longer just tools, but independent players in medical, cognitive and infrastructural change.

At the same time, our understanding of artificial intelligence is changing. With the “Darwin Gödel Machine” approach, systems are emerging that improve themselves, test their own hypotheses and continue to develop beyond fixed architectures - like a digital evolution. This paradigm shift means that AI will no longer be limited to static training data, but will learn through open exploration, similar to biological organisms.

This is nothing less than the beginning of an era of autonomous cognition.

In Today’s Issue:

• Revolutionary Artificial Blood could end shortages worldwide

• Sam Altman on AGI: Why the real milestone is smooth scientific progress

• Near-zero latency teleoperation shows a new era of robotic home assistants

• Solar power surges transform the global electricity game

• And more AI goodness…

All the best,

Scientific breakthrough: artificial blood for all blood groups

An artificial blood product has been developed in Japan that can be used regardless of blood type and can be stored for up to two years without refrigeration. The researchers at Nara Medical University rely on recycled hemoglobin from expired donor blood, which is embedded in a protective lipid membrane. These “blood cells” are virus-free, can be used universally and no longer require blood group matching.

Clinical trials are underway: Healthy volunteers are receiving doses of up to 400 milliliters to test efficacy and safety. Universal blood could soon save lives, particularly in disaster areas, during military operations or in rural clinics. The vision: blood reserves that are available and storable at all times - regardless of donor shortages or logistics problems.

For the AI and biotech community, this project impressively demonstrates how data-driven analyses, automated manufacturing processes and biotechnological innovation work together. Artificial blood products could be the beginning of a new era - one in which healthcare systems are more resilient, more global and fairer.

Why it matters: Artificial blood can end supply shortages and transform emergency medicine worldwide. It makes modern medicine less dependent on blood donors - a real paradigm shift.

Sources:

🔗 https://www.newsweek.com/artificial-blood-japan-all-blood-types-2079654

🔗 https://www.tokyoweekender.com/entertainment/tech-trends/japanese-scientists-develop-artificial-blood

🤖 Huge News in the world of AI Agents

Runner H is the last AI tool you'll ever need! Backed by a historic $220M seed round, H just released their flagship AI agent that lets you automate entire tasks in a few seconds.

Some tasks you can delegate to Runner H:

• Reading your important emails and drafting (or even sending!) replies

• Creating a Google Sheet with trending ad ideas and sharing it with your team on Slack

• Finding job opportunities and applying for you.

Offload some work to Runner H and see how good it feels! Try it now for free at https://runnerh.com

Sam Altman on AGI: Progress Over the Finish Line

Sam Altman emphasizes that the true milestone isn’t the moment we achieve AGI or superintelligence, but the steady, smooth exponential progress toward it. What truly matters is developing systems that can autonomously discover new science or help humans accelerate scientific breakthroughs. According to Altman, if global scientific discovery quadruples, that would meet any meaningful test for AGI.

As expected, Sonnet-4 and Opus-4 are the absolute top dogs in the SWE benchmark.

The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models

The paper “The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models” identifies a key problem when using reinforcement learning (RL) to improve the reasoning capabilities of large language models: the premature loss of entropy, i.e. the diversity of possible decisions. This “entropy collapse” leads to models becoming overconfident too early and thus exploring less, which limits their performance. The authors show that this effect is mathematically predictable and propose two methods - Clip-Cov and KL-Cov - to specifically control entropy. These approaches promote exploration and could drive the development of more powerful, autonomous AI systems.

Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents

The paper “Darwin Gödel Machine” presents an AI system that evolves autonomously by changing its own code and empirically testing the effectiveness of these changes. Inspired by biological evolution and scientific progress, the system generates a growing collection of specialized AI agents that improve each other. In contrast to previous approaches based on fixed architectures, this open, evolutionary process enables continuous self-improvement. This could accelerate the development of autonomous AI systems and reduce dependence on human intervention, which could have far-reaching implications for technology, the economy and society.

MedHELM: Holistic Evaluation of Large Language Models for Medical Tasks

The paper “MedHELM: Holistic Evaluation of Large Language Models for Medical Tasks” presents a new evaluation system that comprehensively tests the capabilities of large language models (LLMs) in real-life medical applications. In contrast to previous tests that focus on standardized exams, MedHELM covers the complexity of everyday medical practice with 121 tasks in five categories - from clinical decision support to patient communication. The results show clear differences in performance between the models and emphasize the need for practical evaluations to ensure the safe use of AI in healthcare.