Just Two Hours! The World's First Fully Automated AI Scientist, Robin, Revolutionizes Traditional Research Paradigms

The non-profit research institution FutureHouse published a significant paper in the prestigious academic journal Nature, officially unveiling the world's first multi-agent AI system capable of achieving "full closed-loop automation" in scientific discovery—Robin. In less than two hours, the system completed an amount of work that typically takes human scientists nearly four months, or about 900 hours.

Like an tireless dream team of researchers

Robin's breakthrough lies in its clever integration of three clearly defined AI agents, connecting the entire process from hypothesis generation to hypothesis validation. The "Crow" agent is responsible for quickly scanning vast amounts of literature and determining experimental strategies, the "Falcon" agent conducts in-depth evaluations, while the "Finch" agent specializes in data analysis, independently writing code, and creating rigorous statistical charts.

In a practical test targeting the blinding eye disease "dry age-related macular degeneration (dAMD)," Robin consumed hundreds of documents in just half an hour. It not only accurately identified the core pathogenic mechanism—the impaired phagocytic function of retinal pigment epithelial cells in the eye, but also precisely screened multiple potential new drugs from a large drug library, significantly shortening the early phase of drug development.

Autonomously cracking pharmacology and opening new directions

After human scientists conducted in vitro cell experiments based on the list of candidates proposed by Robin, they confirmed that the glaucoma clinical drug "Ripasudil" has remarkable potential for repurposing. Even more astonishingly, when facing experimental data, Robin autonomously proposed and designed an RNA sequencing (RNA-seq) plan, successfully deciphering the underlying pharmacological mechanisms of the drug.

During the data analysis process, Robin even uncovered a long-overlooked clue of significant upregulation of the "ABCA1" gene, opening an unexpected new direction for the development of novel targeted therapies for this eye disease. This disruptive efficiency and insight mark the beginning of a completely new era in AI-driven scientific discovery.

Google Cracks Down on AI Poisoning: GEO Spam Content Will Be Devalued and Removed

Google updated its generated search governance guidelines in mid-May, officially classifying malicious manipulation behaviors in Generated Engine Optimization (GEO) as spam content. This move targets the challenge of "AI poisoning" in the era of large models, aiming to maintain the fairness of the search ecosystem by taking a stand against advertisers attempting to deceive "AI" by fabricating information.

Google CEO Sundar Pichai: Frankly Admits Lags in Programming Field, AI Search Transition Will Take a Steady Approach

After the I/O conference, Google CEO Sundar Pichai accepted a podcast interview, openly discussing the AI competition. He admitted that Google leads in overall intelligence and multimodal capabilities, but lags in agent-based programming and complex long-term tasks. The conversation also touched on changes in search patterns and AI safety, reflecting Google's open attitude as it faces challenges in the AI race.

Mianbi Intelligence Jointly Launches China's First 1.58-bit Large Model BitCPM-CANN with Tsinghua University

Mianbi Intelligence, in collaboration with Tsinghua University and the OpenBMB open-source community, released and open-sourced China's first ternary (1.58-bit) large model, BitCPM-CANN, trained on the Huawei Ascend platform. This model achieves a breakthrough in low-bit training, with native development across the full chain from quantization operators to training algorithms. It offers four versions ranging from 0.5B to 8B parameters, showcasing....

Powerful Collaboration! Texas Plans to Build a 2.5 Gigawatt Nuclear and Gas Hybrid Power Plant to Address the Surge in AI Energy Demand

U.S. energy company Blue Energy has partnered with GE Vernova to develop a 2.5 gigawatt hybrid power plant in Texas, incorporating both nuclear and natural gas power facilities to address the increasing electricity demand from AI data centers. The project pioneers the "gas to nuclear" model, using natural gas for quick power supply and gradually transitioning to nuclear energy, effectively addressing the challenges of long approval and construction cycles for traditional nuclear power plants. It provides a flexible and stable power solution for high-energy-consuming industries.