Amazon to use AI-designed material to capture carbon

Currently, data centers consume a lot of energy, use a lot of water, and produce a lot of waste heat. The computation needed for AI at the massive scale projected for the next decade will make all of these negative effects much, much worse - unless something changes. Amazon is trying to lead the way to a greener way of computing, by tasking AI to solve its own environmental problems. This week, Amazon announced a partnership with the AI materials discovery startup Orbital, to pilot deploying an AI-designed carbon-capturing material that locks up carbon dioxide in nanopores, removing it from the ambient air. The captured carbon dioxide can then be stored underground or used in a variety of biotech processes that produce usable products without adding to the planet’s greenhouse gas burden. Orbital is working on other AI-designed materials to deal with the issues of water usage and waste heat production. Amazon hopes to “flip the script” and make its data centers into “green factories” for environmental renewal rather than further climate damage.

Amazon to pilot AI-designed material that sequesters carbon

US issues even more restrictions on AI chip exports to China

In the waning days of the Biden administration, the US Commerce Department has issued a third round of export restrictions to China, this time including chip-making equipment, as well as the advanced chips themselves. Incoming president Trump is expected to uphold or even extend such measures, if his rhetoric is any guide. China predictably vowed to retaliate with trade restrictions and tariffs of their own, targeting vulnerable sectors of the US economy, such as agriculture.

US Commerce Dept. issues even more restrictions on chips for China

Intel CEO gets fired by his Board after a disastrous 3½ years

On Monday, the Board of storied US chipmaker Intel booted out the man they had chosen to lead the company back to glory a mere 3 ½ years ago. Pat Gelsinger, a 25-year veteran of Intel who had left to become the acknowledged “best CEO in tech” at VMware, returned to what he called his “dream job” at the helm of the troubled company, which in better days had supplied the chips that fueled the rise of personal computers. His dream became the shareholders’ nightmare, as the share price slid to less than half of its value before his accession. Intel was a laggard in AI chipmaking, and continued to hemorrhage money, to the point where US policymakers in Washington began seriously discussing a bailout to keep US chipmaking viable, a national security priority in the age of AI. This story affirms the often-quoted maxim of legendary investor Warren Buffet; “When a manager widely regarded as talented joins a company widely regarded to be a bad business, it is the reputation of the business that survives.”

Intel CEO gets fired by Board after a disastrous tenure

Claude releases open-source universal app connector

OpenAI rival Anthropic, whose chatbot is endearingly named Claude, has released an open-source standardized method for connecting AI models to databases and applications, called MCP (for Model Context Protocol, not that anyone will ever remember that abomination of nomenclature.) Name aside, MCP makes it easier to craft “agents”, which are AI systems that can actually do stuff for you, like order you a pizza, or find and reserve a hotel room for your trip to Portland, Oregon. Agents are widely regarded as The Next Big Thing in AI, and having a standard toolkit for connecting to apps and databases may turbocharge their development. Maybe Anthropic could use MCP to connect to a better database of names for software.

Claude releases open-source universal app connector

Music professor claims Suno AI is better at songwriting than 80% of his students - but is far behind the best

Suno, OpenAI’s AI-powered songwriting website, has just gotten a major upgrade, to version 4, and it is causing quite a stir in music circles. A professor at the prestigious Berklee College of Music in Boston claims in a Reddit post that Suno is “better than 80% of my students”, then goes on to add “but my best students beat it by miles. The industry’s best also wins.” He concludes by judging Suno’s creations as “ready-to-eat service music (advertisement/library)”. Perhaps so, but speaking authoritatively as a musical ignoramus and klutz, the ability to create silly songs for your children’s (or grandchildren’s) birthdays and sports triumphs is priceless. Suno may flop commercially, but it is a winner for the amateur family comedian (and we are legion.)

Music professor claims Suno AI is better at songwriting than 80% of his students

“The AI reporter that took my old job just got fired after 2 months”

A local newspaper on the Hawaiian island of Kauai had trouble retaining human reporters, so it paid an Israeli AI company to create 2 AI video chatbots to deliver the news on YouTube, Facebook, and Instagram. The results were either tragic or hilarious, depending on your point of view. The 2 video chatbots, an Asian-looking middle-aged man called “James” and a younger redheaded woman called “Rose”, managed to instantly alienate almost all viewers. They were, well, robotic. They discussed every story - from a fall pumpkin giveaway to a vigil for a labor massacre - in the same distant, matter of fact tone, that showed no empathy or understanding of human emotions. They mispronounced local place names. James’ hands vibrated unnervingly. Within 2 months, the newspaper pulled the plug on the video chatbots, calling the experiment “a success” despite the almost universal negative tone of the comments from viewers. The local newspaper is part of a media conglomerate with more than 100 local affiliates, so an upgraded James and Rose may find their way to a struggling local newspaper near you.

The AI reporter that took my old job just got fired after 2 months

US DOE offers $30 million for AI to break grid connection logjam

The US Department of Energy has allocated $30 million to encourage AI solutions for speeding up approvals for connecting additional sources of clean electricity to the grid. Currently, the interconnection approvals process can take 7 years or more, and there is twice as much clean energy being produced in active but unconnected sites as in grid-connected sites. Connection requires navigating a maze of federal, state, and local laws and regulations, plus securing agreements with landowners, power purchasers, and equipment suppliers. Since AI is very good at reading texts (like laws, regulations, and contracts), and for deciphering complex patterns in the data, the hope is that it can help cut the Gordian knot of impediments to bringing more clean energy into the grid.

US Dept. of Energy offers $30 million for AI solutions for grid connections

Over half of longer LinkedIn posts are AI-generated

LinkedIn, the business social media site, has recently made a big push to include AI capabilities into its site. This includes giving free “help” to users in generating their posts. Free is everyone’s favorite price, and easy is everyone’s preferred level of difficulty for tedious tasks, so the adoption of AI writing “help” has been swift and massive. A recent study has found that over 50% of longer posts (over 100 words) on LinkedIn are likely generated by AI. Since most longer posts on LinkedIn are basically self-advertisements for the author (Look at Meee! I’m so smart!!!), it is no surprise that AI, which is designed to be glib and persuasive, and inherently tends to play fast and loose with the facts, has come to become the preferred co-author on this site where self-promotion is practically the whole point.

Over half of longer LinkedIn posts are AI generated

MIT AI can diagnose pneumonia, but knows when to call in the radiologist

Much of the controversy over using AI in medicine revolves around ceding control over potentially life-and-death decisions to a robot. MIT researchers have developed a system for detecting pneumonia on Chest X-rays that explicitly keeps a human in the loop on critical decisions. The AI system will first use its pattern-deciphering capabilities to classify a chest X-ray as showing pneumonia or not. On tough calls, or on readings that suggest a need for medical intervention, the chest X-ray and all relevant medical records are forwarded to the expert radiologist for definitive decision-making. The authors write that this hybrid human-AI system outperforms both the AI-only and the human-only comparison workflows.

MIT AI can diagnose pneumonia, but knows when to call the radiologist

NIH develops AI to match volunteers to clinical trials

Researchers at the National Institutes for Health have developed an AI system that helps match patients to potentially appropriate clinical trials. Given the large numbers of clinical trials open at any given time, and their varied inclusion criteria, matching patients to the appropriate clinical trial by hand has proven to be complex, time-consuming, and error prone. The AI system simplifies and partially automates this process, which it is hoped will recruit patients to appropriate clinical trials faster, advancing scientific discovery, while simultaneously getting patients the care they need.

NIH develops AI to match volunteers to clinical trials

AI better at diagnosing skin cancer than experienced Dermatologists

Researchers in Germany, France, and the US trained an AI system on over 100,000 labeled images of both malignant melanomas and benign moles. At the end of training, the AI’s performance was compared to that of a panel of 58 experienced Dermatologists from 17 countries around the world, and it was found that the AI missed fewer melanomas (i.e. had higher sensitivity) and misclassified fewer benign moles (had higher specificity) than the human physicians. This disparity held true even when the Dermatologists were given clinical information about the patients, which did improve their results. Still, the AI, which considered only image analysis without clinical information, continued to be superior in diagnosis of the lesions.

AI better at diagnosing skin cancer than experienced Dermatologists

Can AI decrease the cost of care by reducing administrative burden?

Why is US health care so expensive? Our cost per resident is nearly twice the average of other developed countries, while our health outcomes are generally worse. One part of the answer is clearly because the US tends to practice a more technology-intensive style of medicine (which, paradoxically, doesn’t seem to improve our outcomes.) Another part of the equation is that our health care system is not really a system, and overlapping administrative regimes of payers and providers constitute a complex maze for patients to navigate. Since 1975, the number of administrators in the US health care system has ballooned by 32 times, while the number of practicing physicians per 1,000 population has roughly doubled for a population that is significantly older and more burdened by chronic disease than was the case 50 years ago. Much of administrative work is paperwork, and AI can be very, very good at paperwork. So, one approach to solving the health care cost dilemma may be to stem administrative bloat by letting AI push the paper.

Can AI decrease the cost of health care by reducing administrative burden?