China’s “Manus” AI Agent creates another “DeepSeek moment”

China stuns the AI world yet again. Just like January’s flurry over DeepSeek, now another Chinese AI startup (this one called Butterfly Effect, get it?) from out of nowhere and demo-ed an eyepopping new product that puts the heat to the huge US-based AI incumbents like OpenAI, Google, Anthropic, Facebook/Meta, and others. This time the product is called “Manus” (Latin for “hand”) and it is a system of AI agents that, if the demo is to be believed, can perform multiple real-world tasks simultaneously, like creating a website, reserving a restaurant table, and so on. A number of techies have looked under the hood, and like DeepSeek, the product is built on a lot of US technology very cleverly engineered together to achieve state of the art results. The major US tech in Manus is Anthropic’s Claude AI model, along with an open-source program called “Browser Use” that allows an AI model to take over the user’s computer to perform tasks. The magic is in the integration, and here Butterfly Effect engineers excelled. And Like DeepSeek, Butterfly Effect has apparently mastered the art of going viral on the internet, no small advantage in the modern “attention economy.”

China’s Manus AI agent creates another DeepSeek moment.

OpenAI plans to sell AI agents for $2,000 to $20,000 per month

OpenAI is reliably reported to be planning to rent out AI agents for a monthly fee ranging from $2,000/month to $20,000/month. These AI agents are envisioned to serve as, variously, a personal assistant for a high-income knowledge worker for $2,000 per month, a high-end software developer for $10,000 per month, and a PhD-level AI researcher for $20,000 per month. SoftBank, one of OpenAI’s major investors, has committed to spending $3 billion on OpenAI’s agent products this year alone.

OpenAI plans to sell AI agents for $2000 to $20,000 per month

Musk’s xAI expands massively in Memphis and Atlanta

Elon Musk is not a well man, but he gets stuff done. His AI company, xAI, is now expanding its massive “Colossus” data center in Memphis, Tennessee with an additional 1 million square foot building. The current facility uses 250 megawatts of electricity, and the proposed expansion will bring the total projected power consumption to 1. 2 gigawatts, 5 times as much. He is simultaneously fitting out a data center in Atlanta with $700 million in computer equipment to be shared between xAI and Elon’s social media site, X (formerly known as Twitter.) The man clearly suffers from an Edifice Complex (along with other ailments.)

xAI expands massively in Memphis and Atlanta

Microsoft announces Dragon Copilot for medical care

Microsoft, through its wholly-owned subsidiary Nuance, is the world leader in medical dictations systems, and one of the leaders in the new and fast-growing field of AI Scribes, in which AI listens to the doctor-patient interaction and automatically produces a draft note for the patient chart for the physician to edit and sign. AI Scribes are taking clinical medicine by storm, and now Microsoft is unveiling its next clinical offering, called Dragon Copilot. Building on its Copilot AI workflow integration, Microsoft is extending its Dragon AI Scribe to include language translation, order entry, and more, including clinical decision support in collaboration with Wolters Kluwer Health’s industry-leading tool UpToDate.

Microsoft announces Dragon Copilot for medical care

Christie’s first AI art auction exceeds expectations

Premier auction house Christie’s braved a wave of criticism from artists when it scheduled its first-ever auction of AI-generated works of art, but the results of the auction brought Christie’s only smiles. Nearly 6,500 artists signed an open letter demanding that Christie’s cancel the show, called Augmented Intelligence, because AI models have been trained on copyrighted work without the artists’ permission. Yet the show went on, and brought in over $700,000, with many lots sold above their high estimates. The market for AI-generated art will likely only grow from here.

McDonald’s undergoes an AI makeover

McDonald’s is overhauling all 43,000 of its restaurants with AI systems designed to cut customer wait times and improve the customer experience. AI upgrades include internet-connected kitchen equipment that gives central AI systems data needed to schedule preventive maintenance to avoid downtime, AI-powered voice-interactive drive-throughs, and AI management tools for the restaurant manager. They are also exploring AI vision systems that can visually check the correctness of an order before it is given to the customer. The massive AI overhaul is part of McDonald’s goal of increasing its base of repeat customers from its current 175 million customers to 250 million customers by the end of 2027 (43% growth), and to grow revenue from $25.5 billion to $45 billion (76% growth, so increasing annual revenue per customer by 24%). One wonders whether this massive automation will be accompanied by more centralization in what is currently a franchise-dominated business, and more automation to reduce labor costs of line jobs and to reduce layers of management in field operations.

McDonald’s undergoes an AI makeover

AI experiment-designing system discovers novel physics

Quantum physics is mind-bendingly complex, and designing experiments for testing theories on quantum physics is equally difficult. So scientists in China and at the Max Planck Institute in Germany developed an AI system to help them design experiments on quantum entanglement, a quantum phenomenon in which two photons stay in sync even when separated by vast distances. They fed a description of the purpose of their experiment into their AI, expecting a variation of the standard 4-photon protocol for entanglement. The AI produced a much simpler 2-photon solution, which the experimenters thought was an error. They kept feeding the AI with rephrasings of their experimental goals, and kept getting back the 2-photon design. Finally, they ran the experiment as designed by the AI, and to their very great surprise, it produced entangled photons just as the AI had predicted. Apparently, in learning how to design quantum experiments, the AI learned the underlying physics, better than the human quantum physicists. 

AI experiment-designing system discovers new physics

AI helps vineyards produce more and better wine

Agriculture is not generally thought of as a high-tech industry, but in the age of AI, it is becoming more so. Even wine-growing, one of the most-romanticized sectors in farming, is beginning to use AI to increase crop yields and improve grape quality at harvest. Autonomous, GPS-guided tractors both tend the crops and collect vast amounts of data on weather, soil conditions, and plant health down to each individual vine. AI-controlled dispensers of fertilizer and pest-control chemicals mounted on these autonomous tractors deliver the right amount of each liquid to each plant. Central AI systems take the data from the field operations of these autonomous tractors and calculate which irrigation valves to open, and which to close, based on hyperlocal soil moisture conditions. These central AI systems also continuously calculate projected crop yield for the season. More advanced vineyards are starting to use drones rather than tractors, but current drone technology has not yet been made as user friendly as John Deere and other major tractor manufacturers have made the robo-tractors, so wider adoption of drones likely awaits a user interface upgrade.

AI helps vineyards make more and better wine

Humanoid robots walk police beats in Shenzen, China

Recently, advanced humanoid robots have been seen on police patrols in the city of Shenzen, China. Although currently more of a publicity stunt for both the police and the manufacturer, the demonstrations are aligned with the Chinese government’s push to integrate robots into all sectors of society, including law enforcement. China, like Japan and South Korea, is facing a shrinking population because birth rates have fallen and immigration is tightly controlled. This has caused all 3 societies to turn to robots as a way of maintaining or even expanding the workforce while the population ages and shrinks. Expect more Robocops in the future.

Humanoid robots walk police beats in Shenzen, China

Japan’s AIREC robots gently assist the elderly and disabled

Researchers at Waseda University in Tokyo are developing an AI-enabled humanoid robot to give physical care to elderly and disabled individuals, such as turning, lifting, and helping them pivot from bed to wheelchair and back. This is demanding physical labor for humans, but industrial robots are not flexible or adaptive enough to provide the needed assistance without risk of injury to the person being helped. Waseda University scientists are developing an AI-controlled version of a humanoid robot, dubbed AIREC (AI-driven Robot for Embrace and Care), which adaptively controls the force of the turning and lifting assistance to be stable but gentle. The team involved predict that robots of this type could be in wide use by 2030.

Japan’s AIREC robots gently assist the elderly and disabled

See the video of a simulated assistance

Stanford AI finds natural protein that causes weight loss like Ozempic

Stanford researchers have developed an AI system they call the “Peptide Predictor”, which has identified a naturally-occurring small protein (peptide) that has the same effects as Ozempic in decreasing appetite and encouraging weight loss, but apparently without the negative side effects of the current medications, such as nausea, constipation, and loss of muscle mass. The newly-discovered molecule, dubbed BRP, apparently stimulates different brain pathways than Ozempic to achieve its effects, which may explain its more benign side effect profile. So far, injections of BRP have only been tested in mice and minipigs, but human trials are being planned.

Stanford AI finds natural protein that causes weight loss like Ozempic

AI analyzes brain waves to predict dementia years before symptoms

Researchers at Boston’s Massachusetts General Brigham hospital system have developed an AI system that analyzes the electroencephalography (EEG) tracings of sleeping patients in order to predict future cognitive decline. The study started with the EEG tracings of 281 cognitively normal women over age 65 who had been assessed in previous research on fracture risk. The women were called back to be reassessed 5 years later for this new study, at which point 96 of the women were found to have developed cognitive deficits. The AI system was able to detect subtle abnormalities in brain wave tracings from the original study, particularly in so-called gamma waves that are associated with learning and cognition, that could predict later cognitive decline. The authors foresee a role for wearable home EEG devices that would permit early detection of possible future cognitive impairment, which could allow early intervention with medication and lifestyle changes to prevent or slow declines in cognition.

AI analyzes brain waves to predict dementia years before symptoms