SpaceX uses AI to recapture a rocket… and more

Even Elon’s haters (including - ahem - moi) have to admit - the dude gets stuff done. In an epic feat of engineering not seen since the glory days of NASA 50 years ago, SpaceX used AI-assisted navigation to return a rocket booster back to its launch tower, where giant mechanical arms (nicknamed Mechazilla) gently plucked it out of the sky to let it rest. The feat has been likened to snatching a fly out of the air with chopsticks, Karate-Kid style. In an unusually busy week (even for him) Elon also threw an event called We, Robot to showcase his long-promised self-driving cars (still not ready) and his Optimus humanoid robots (also not yet ready.) The point he was making is that Tesla is not just a car company, it is an AI robotics company that can apply its AI to any mechanical system, from cars to rockets to humanoid workers. Point taken.

SpaceX uses AI to recapture a rocket on its launch tower

Google commits to build small nukes to power data centers

AI consumes a lot of electricity, and all the big players are scrambling to secure sources of power for their multi-billion-dollar data center building sprees. The urgency of the need, combined with a commitment to lower their carbon emissions, has made the companies warm to the idea of nuclear power, long a pariah among energy sources due to highly publicized disasters at Three Mile Island in the US and Fukushima in Japan. (Microsoft has already committed to reviving Three Mile Island, as we have reported previously.) Now Google has signed an agreement with Kairos Power to build a series of small nuclear power plants near its data centers, totaling 500 megawatts (enough to power approximately 500,000 homes) by 2035. The new nukes are reportedly smaller and safer than the old nukes (salt-cooled, rather water-cooled, reducing chances of blowouts or meltdowns), and are likely to get regulatory approval. The US already has the most operational nuclear power plants in the world, but nuclear only supplies around 20% of our electrical power (in France, it is nearly 70%.)

Google commits to build small nukes to power data centers

Nvidia announces first AI Tech Community in Pittsburgh

Leading AI chipmaker Nvidia, now one of the three most valuable public companies in the world after what founder-CEO Jensen Huang calls 20 years of “pain and suffering”, is not resting on its laurels. It is planting seeds for its future success by funding local AI research and development collaboratives, involving universities, startups, and other local organizations in a technology ecosystem modeled on Silicon Valley. Nvidia will fund its first AI Tech Community, as the initiative is called, in Pittsburgh, with joint research centers at Carnegie-Mellon University (focused on robotics) and the University of Pittsburgh (focused on health care.) Local AI startups will also be part of the collaborative, in an effort to quickly bring lab results to the market. The initiative displays Huang’s signature combination of boldness, creativity, breathtaking ambition, and hands-on practicality. Say what you will about Jensen Huang - he ain’t dull.

Nvidia announces its first AI technology collaborative

Amazon’s new warehouses will use 10x as many robots

Amazon announced last week that its next-generation fulfillment centers will use 10 times as many robots as its current warehouses. The first of the Amazon’s new breed of fulfillment centers is located in Shreveport, Louisiana; it contains 3 million square feet of warehouse space on 5 floors, the equivalent of 55 football fields. The cavernous interior is crisscrossed by bevies of stocking and picking robots. Despite the high level of automation, the warehouse is projected to employ 2,500 humans as well.

Amazon’s new warehouses will use 10x as many robots

Air Street Capital releases its annual State of AI report for 2024

Each fall, Air Street Capital summarizes the major trends in AI for the past 12 months. The link below leads to the full 211-slide report. Here are the top 10 takeaways, in no particular order:

1. All major AI models are roughly equivalent in power, including the open-sourced.

2. Small models are getting ever more powerful, and will run on your own device.

3. The next AI frontier is reasoning, planning, and AI agents that do work for you.

4. Big players are turning to nukes to power their plethora of new data centers.

5. AI is already contributing to science - as evidenced by 2 Nobel Prizes this year.

6. Robots are coming soon - humanoid, 4-legged, and aerial drones.

7. Self-driving cars are a commercial reality today.

8. AI is already helping write computer code for developers and users alike.

9. Deepfake technology is advancing rapidly with new video and voice-cloning tools.

10. Governments are struggling to write sensible regulations that won’t kill innovation.

State of AI report for 2024

Penn State invents an AI-powered electronic tongue

Researchers at Pennsylvania State University have developed an electronic tongue which uses AI to distinguish characteristics of liquids. The electronic tongue can, for example, detect subtle signs of spoilage, important in food safety and production, and can also detect trace contaminants in biological fluids, important in medical diagnostics and therapeutics. Interestingly, the electronic tongue performed better when the AI developed its own classification scheme for ingredients rather than relying on the scheme supplied by the researchers.

Penn State invents an AI-powered electronic tongue

Nature article

Wimbledon replaces all 300 line judges with AI

The venerable All England Lawn Tennis Club has announced plans to scrap a 147-year tradition and use AI to replace all 300 of its line judges at its annual Wimbledon Championships, the oldest and most prestigious tennis tournament in the world. The AI line-calling system utilizes 12 cameras strategically placed on every court, as well as microphones to detect the sound of the ball. All this real-time information is transmitted to a computer in a control room that runs the AI model that makes the final call with human supervision. Wimbledon spokespeople say that the system was extensively tested during this year’s championship matches, and passed all tests with flying colors.

Wimbledon replaces all 300 line judges with AI

Dario Amodei’s “Machines of Loving Grace”

Dario Amodei, CEO of AI startup Anthropic, arch-rival to OpenAI, has just penned a long meditation on the implications of inventing AI that is smarter than any human, or maybe smarter than all of us combined. He says that this event could be as near as 2 years from now. The heart of his essay is devoted to laying out his assessment of the consequences of such a leap in machine intelligence. Briefly, he expects that scientific progress will accelerate 10-fold, so that the following 5-10 years will see 50-100 years worth of progress. This will extend human life, enable widespread prosperity, force a restructuring of work and society, and set off a fateful competition between autocracies and democracies.

Whatever the merits of Amodei’s analysis, it is reminiscent of another recent techno-utopian essay, OpenAI CEO Sam Altman’s “The Intelligence Age.” Sam soon after raised $6.6 billion in investment for OpenAI, the largest venture capital raise in history, and proof of the adage “When you can’t show profits, sell investors on the dream.” Amosei may be heading for another capital raise at Anthropic, and techn-utopianism may be just the pixie dust to get investors over the line.

Dario Amodei’s “Machines of Loving Grace”

AI gets its 2nd Nobel - this time in (bio)Chemistry

Merely one day after the Swedish Royal Academy awarded the Nobel Prize in Physics to two founders of the field of AI, they awarded the Nobel Prize in Chemistry to 3 pioneers in computational biology, including Demis Hassabis, the CEO of Google’s DeepMind AI research arm. Hassabis’ DeepMind team was recognized for creating the AlphaFold AI system that cracked a 50-year-old scientific problem of predicting a protein’s shape solely from the sequence of its amino acids. The AlphaFold system and its outputs have been made open source, and these have already sped scientific research in a variety of areas in biology.

AI gets its 2nd Nobel - this time in (bio)Chemistry

Stanford Medical Center has implemented 30 AI initiatives

AI promises a revolution in a number of fields, but nowhere so much as in medicine. “Ambient documentation”, the practice of using AI to create a transcript of a physician-patient interaction, then prepare a summary for the physician to edit and put into the medical record, is quickly becoming the standard of care for large health systems, and progress is reported almost daily in AI applications to diagnostics. Stanford Medical Center is not only a major health delivery system, it is also located in the heart of Silicon Valley, and so is nearby many of the companies and research labs that are driving progress in AI. A recent report notes that Stanford Medical Center is currently engaged in 30 AI projects for clinical medicine.

Stanford Medical Center has implemented 30 AI initiatives

AI discovers 160,000 unknown viruses

An AI system known as LucaProt has been used by an international team of researchers to identify fragmentary RNA sequences recorded in public databases and assign them to species. This has resulted in the identification of over 160,000 new species of virus, a feat that would have taken many years using conventional procedures.

AI discovers 160,000 unknown viruses

Abridge incorporates UpToDate into its clinical decision support

Medical AI startup Abridge, one of the leading “ambient documentation” vendors in the US, has signed an agreement with Wolters Kluwer, publisher of the UpToDate clinical information system for physicians. Abridge will incorporate UpToDate into its clinical decision support functions. This gives Abridge access to UpToDate’s industry-leading best practices guidelines, while allowing Wolters Kluwer to freshen up its somewhat dated-looking search interface with a breath of AI sizzle.

Abridge incorporates UpToDate into its clinical decision support