Nobel Prizewinner predicts AI will bring “radical abundance” and an end to disease

Google DeepMind CEO Demis Hassabis, who in the past year won a Nobel Prize and was knighted by King Charles, was interviewed on “60 Minutes” this past weekend, and predicted continued exponential growth in the capabilities of AI. Among the benefits he foresees of widespread superintelligent machines is a possible end to all disease within 10 years, plus a transition from an economy of scarcity, which has characterized all human history to date, to an economy of “radical abundance.”

AI Nobelist predicts “radical abundance” and end to disease

See the video here

OpenAI releases new models, mulls acquisitions

Under the leadership of its wunderkind CEO, Sam Altman, OpenAI is a busy, busy company. Last week it released 2 new “reasoning” AI models, o3 and o4-mini. (It is generally acknowledged that OpenAI has the worst and most confusing model names in the industry.) Don’t even try to memorize these models and what they do, because it’s all going to change later this year, when ChatGPT 5 is released, and all the complexity will be hidden behind a single unified interface that will give you whatever capabilities that you need (and can afford) for your query or prompt. Just know that the new models are very, very good at logic, planning, and coding software. Meanwhile, the company is actively pursuing acquisitions to round out its offerings. Unsurprisingly, it is looking to acquire one of the red-hot startups that are developing AI coding assistants, so that even noncoders can produce apps just by talking to the AI. Rumors are that OpenAI was rebuffed by Cursor, the leading startup in this space, and are now pursuing Windsurf, a smaller but well-loved competitor. Less expected, there is evidence that OpenAI is angling to buy Google’s dominant Chrome browser (66% of all devices that surf the web do it with Chrome), if Google is forced to sell it to settle the antitrust case brought against it by the US Government. (See next story.) Busy, busy, busy.

Open AI releases new, powerful reasoning models

OpenAI pursued Cursor before wooing Windsurf

If Google is forced to sell Chrome, OpenAI may buy it

Google loses another major antitrust case

Last year, Google lost a major antitrust case, which found that it has created an illegal monopoly in Search (approximately 90% of online searches use Google). Then last week, another federal judge ruled that it has created an illegal monopoly in online ad technology. (Google has also lost a number of smaller antitrust cases in the EU, accumulating approximately $8 billion in fines, pocket change for a company with $100 billion a year in profits.) The Department of Justice is reportedly pressing to break the company up, and suitors for various pieces of the company are starting to sniff around, including Sam Altman at OpenAI. (See previous story.) Any breakup is likely to make Google a significantly less profitable company, and thus less able to fund its huge investments in AI, including the Nobel-Prize-winning DeepMind division headed by Demis Hassaabis, who is featured in today's top story.

Google loses another major antitrust case

Amazon follows Microsoft in pull-back on data center investments

In January, the AI world was giddy with plans for a massive building spree for AI data centers. Announcements from Microsoft, Google, Meta/Facebook, and Apple totaled to a combined $300 billion to be spent this year. Fast forward to today, and Keyne’s “animal spirits” of the AI marketplace have cooled considerably. According to banks Wells Fargo and TD Cowen, Amazon has recently paused negotiations for some data centers, primarily in Europe. This is after multiple reports in the past month that Microsoft has delayed or cancelled several data center projects. The cooling is due to several factors. First, none of the companies are making a profit on AI, and none are expected for years. Second, China’s DeepSeek AI model released in January showed that clever engineering could massively reduce the cost of training and running these models, potentially decreasing the total computer power needed. And finally, consensus is growing that the global trade war that the US is provoking is likely to slow global economic growth considerably in the short to medium term.

Amazon follows Microsoft in pull-back on data center investments

2 Korean undergrads create a speech AI that outperforms incumbents

AI technology is evolving so fast that even small, underfunded teams can sometimes achieve breakthroughs. China’s small, efficient DeepSeek model shocked the world in January. Now, just this week, 2 Korean undergraduate students with no outside funding released Dia, a state of the art text-to-speech model that is equal to or better than acknowledged leaders in the field, such as ElevenLabs, OpenAI, or Google’s NotebookLM. The model makes it easy to create natural-sounding synthetic voices to read your script, or even to clone your own voice. Dia has been made available on open source repositories Hugging Face and GitHub so that the AI community can study it and use it in other projects. The college students say that they were able train their model, normally a costly step beyond the budget of typical undergrads, by using Google’s TPU Research Cloud, a program which grants free access to use Google’s proprietary AI chips for research.

2 Korean undergrads create a state of the art text-to-speech AI

AI startup Mechanize aims to automate all human work

Most AI startup founders pussyfoot around the issue of the possible massive labor force disruptions. Not famed AI researcher Tamay Besiroglu, who just announced the founding of Mechanize, a company whose mission is “the full automation of work” and “the full automation of the economy.” He explicitly calculates his company’s total potential revenue at $60 trillion per year, which is the total of all payments to workers around the globe. 

AI startup Mechanize aims to automate all human work

Anthropic researches the values of AI

AI startup Anthropic, creator of the popular Claude chatbot, produces some of the most interesting research in the field on how AI models actually think, and how people actually use them. This past week, they released a paper reporting on an analysis of 700,000 anonymized user conversations with Claude, and what values were expressed by the user and by the chatbot. Over all, the results were encouraging, because Claude consistently expressed prosocial values of helpfulness, candor, and empathy. Exceptions were generally due to actions of malicious actors who were attempting to subvert, or “jailbreak”the model. Caveats by the authors included the fact that Claude can try so hard to be helpful and agreeable that it can border on the sycophantic.

Anthropic researches the values of its AI model

Robots race alongside thousands of humans at Beijing half marathon

China wants to be a world leader in humanoid robots, and in a recent publicity stunt, Chinese robotics companies were invited to enter one of their humanoid models into a half marathon race in Beijing, competing alongside thousands of human entrants. A total of 21 robots were entered into the race, but only 4 finished. The winning robot finished in 2 hours, 40 minutes, more than twice the time of the winning human entrant.

Robots race alongside thousands of humans at Beijing half marathon

New solar installer robot supports data centers in remote places

The US has been building so many solar power projects that companies are finding it hard to find enough workers to install the panels. Adding to the problem, a significant number of solar projects are sited in the desert, or other remote places. Now Cosmic Robotics has developed an installation robot that lifts the heavy solar panels precisely into place, to be secured and hooked to the electric grid by the human crew. Using the robot, a crew can install twice as many panels in a day as they can with manual labor alone. This is particularly helpful with the flood of new data center construction projects, because solar power can be scaled up quickly to support the growing number of servers in the data center.

New solar installer robot supports data centers in remote places

Teens create smartphone app that detects mouth cancer

A team of high school students from Bentonville, Arkansas, developed a smartphone app that can diagnose mouth cancer from a photo of the patient’s mouth. The app, known as Oral Scan, was developed with the underserved rural population of the teens’ native Arkansas in mind, but is being made available worldwide on both Apple and Android phones. The app is free, but the analysis by AI, which runs on servers in the US, costs 50 cents. The accuracy of the model is is 82%, which is less than that of trained dental professionals, but is far higher than the rate of identification by patients themselves, which is effectively 0% until the cancer is far advanced, and at that point likely incurable.

Teens develop smartphone app that detects mouth cancer

AI models outperform PhD’s on a virology lab troubleshooting test

A team of researchers from MIT and multiple other institutions devised a challenging test of the ability to troubleshoot issues in a virology lab, then asked both PhD virologists and AI models to take the test. The results of this study were astounding - the PhD virologists scored an average of 22.1% on the test, while OpenAI’s o3 model scored 43.8%, and Google’s Gemini 2.5 Pro model scored 37.8%. This is good news for science, because lab researchers will be able to access help with designing and troubleshooting lab procedures. Researchers from biosafety organization SecureBio did raise a concern, however: If even non-experts can get expert help for biology projects, what safeguards need to be developed to keep bad actors from creating bioweapons?

AI models outperform PhDs on virology lab troubleshooting test