Taiwan farmers battle AI chips for water

AI chipmakers’ massive water demands are squeezing out Taiwan farmers amid a devastating drought spurred by climate change. Taiwan’s traditional rice farming methods use copious amounts of water to flood their fields. In former years, seasonal monsoons drenched the landscape with plenty of water for all. Since 2021 though, monsoons have largely disappeared in Taiwan due to the planet’s shifting climate, and now the local farmers need allocations from the government-controlled water supply. The Taiwanese government’s solution so far has been to allocate water to the chipmakers whose exports fuel the local economy, and pay the farmers not to grow rice. See below for a picture of the sunbaked earth where rice paddies used to flourish.

Taiwan farmers battle AI chipmakers for water

Meta releases Llama 3.1, world’s largest open source AI model

While OpenAI, Anthropic, and Google battle for AI supremacy with their proprietary models, Mark Zuckerberg’s AI shop keeps pumping out leading edge models and making them available for free. Crazy? Yeah, like a fox. Zuckerberg’s social media empire makes enough money that he doesn’t have to charge for AI the way AI-only companies like OpenAI and Anthropic must. And his revenue is not directly threatened by AI the way Google’s ad revenue is if AI radically demonetizes search. So Zuck plans on stealing a tactic from Google’s past when it gave the Android operating system away for free to make sure that there was an alternative to Apple’s IOS. If a whole application ecosystem evolves around Llama models, Zuck is sure to find a way to profit. Meanwhile, he is pushing Llama-fueled AI assistants into all of his social media properties, which with their billions of users, could soon make Llama the most-used AI model in the world, eclipsing ChatGPT.

Meta releases Llama 3.1, world’s largest open source AI model

Microsoft develops SpreadsheetLLM

Massive amounts of corporate data are tied up in spreadsheets, but AI large language models like ChatGPT do not deal with spreadsheet data well. Now Microsoft has developed SpreadsheetLLM, a system that encodes and compresses the data in spreadsheets in a way that makes it easy for AI models to use. This may not sound like a big deal, but given the truly ginormous amounts of corporate data locked away in spreadsheets, it is truly a BFD.

Microsoft develops SpreadsheetLLM

Ukraine scrambles to develop AI-enabled war drones

Drones and other autonomous war vehicles have been crucial in Ukraine’s battle against Russian invaders (see our previous reporting.) Now Ukraine is breeding a bevy of startups to take its drone game to the next level, making AI-enabled war drones that can autonomously coordinate in swarms that can take on targets previously impractical to attack. Now, instead of having to have a human operator for each drone, one human can command a fleet of 20 or more synchronizing drones and give the attack command to all at once.

Ukraine scrambles to develop AI-enabled war drones

Educational startup will develop AI tutors

Legendary AI pioneer Andrej Karpathy, who headed AI efforts at both Tesla and OpenAI, has a new project. He has announced Eureka Labs, a startup that plans to transform education by developing AI teaching assistants that would allow “anyone to learn anything.” Details are scanty, but Karpathy’s track record and his known passion for teaching (his introductory AI course on YouTube is a must for serious AI novices) are stirring excitement in both tech and educational circles.

Educational startup will develop AI tutors

Google’s laptop AI climate model can beat supercomputers

Current weather forecasting is based on complex physics models that require a supercomputer to solve. Now Google has developed an AI model that is just as accurate, but requires just a fraction of the computing power (1000 to 100,000 times less, to be more precise.)

Google’s laptop AI climate model can beat supercomputers

AI model generates realistic street views from text

Researchers from Stanford and Google have developed Streetscapes, an AI model that can generate realistic-looking videos of street views from a text prompt. The image below illustrates that the model output can generate street views consistent with the architecture of a given city, as seen in any given weather, at any time of day. The link takes you to the project page, where you can view multiple short clips of the capabilities of the system.

AI model generates realistic street views from text

AI is already taking jobs in the videogame industry

Videogames are a $200 billion a year industry (4 times the size of movies and video), and are ripe for an AI revolution. Gamers are already used to not-quite-realistic characters speaking wooden dialog in fantasy settings, so the bar for AI is lower than in live-action sectors. It appears that the major game studios have gotten the AI bug, and are already slashing headcount as they find ways to infuse AI into their offerings. Credible sources estimate that studios have laid off or failed to rehire more than 11,000 positions over the past year. Employees at game studios are largely nonunion, so there has been no organized pushback as there was in last summer’s SAG-AFTRA actors and writers strikes, which forced the studios into a historic contract with protections against replacement by AI without compensation. At this point, it seems that the entire videogame industry is likely to be transformed by AI in the not-so-distant future.

GPs in the UK use AI to increase cancer detection rates

England’s National Health Service has launched a program that uses AI to scan patient records and alert GPs (general practitioners, the primary care physicians in the UK system) of cancer risks in their patients. The system also gives the physician suggestions on appropriate tests or scans to order to work up the potential cancer. A study of the impact of the program on the patients of the first 35 practices to adopt the system found the cancer detection rate at the primary care level increased from 58.7% to 66.0%, a 12% improvement over baseline.

GPs in the UK use AI to increase cancer detection rates

VA awards contracts for AI scribes to reduce MD burnout

The U.S Department of Veterans Affairs has awarded contracts to leading AI scribe vendors Abridge and Nuance for pilot programs to test their AI scribe systems in real-world clinical care at several VA healthcare facilities. The goal of the pilot program is to test the efficacy and reliability of the systems, while also gauging their potential for reducing physician burnout. Multiple studies have shown that the administrative burden of documentation is a major factor in physician burnout, and AI scribe systems are increasingly being deployed to deal with this problem.

VA awards contracts for AI scribes to reduce MD burnout

UCLA AI system slashes prostate cancer error rates by 50%

Surgery for prostate cancer often cause incontinence or erectile dysfunction, so there is a trend nowadays to attempt minimally-invasive “focal” treatment, which leaves as much of the prostate intact as possible. Precision is a virtue in such treatments, because leaving cancer behind can ultimately be fatal for the patient, but taking out too much risks the poor outcomes that focal treatment was designed to avoid. Researchers at UCLA have developed an AI model which can accurately define an appropriate resection margin, which in testing reduced errors in setting the margin by half.

UCLA AI system slashes prostate cancer error rates by 50%

MIT develops AI system to improve breast cancer detection

Scientists at MIT and ETH Zurich have developed an AI system that classifies breast cancer biopsy cells into 6 different categories, which more precisely predicts the risk of tumor invasion and dissemination than current standard methods. This 6-tier classification scheme spontaneously arose from the model as it was being exposed to the training data.

MIT develops AI system to improve breast cancer detection