How people actually use AI now, versus 2024

AI tools continue to evolve rapidly, and so does how people actually use them. Harvard Business Review published a deeply researched article last year on how people are actually using generative AI, and this year they have updated this research, allowing for a comparison of the the shifts in usage in just a single year. Much has changed. Of the top 100 use cases for AI in 2025, 38 are new this year, as are 2 of the top 3 uses. All 3 of the top uses of AI in 2025 relate to personal and professional support: Therapy/Companionship at number one, Organizing My Life in the second spot, and Finding Purpose third. This exemplifies an overall shift in the list to the personal and emotional, counterbalanced somewhat by much more use of AI in generating and improving software by professional coders. Somewhat surprisingly, people are increasingly using AI as a companion, therapist, coach, and tutor, overshadowing somewhat more utilitarian uses such as editing text. Given the rapid advances in AI capabilities, which show no signs of slowing down, it seems likely that new use cases for AI will continue to emerge.

HBR research on how people actually use the internet, last year vs. this year

Nvidia to invest $500 billion to bring its chipmaking back to the US

Premier AI chipmaker Nvidia has announced that it will invest up to $500 billion over the next 4 years on building AI servers in the US, bringing back its chip manufacturing from Taiwan. Nvidia is partnering with the world’s leading chip manufacturer, TSMC of Taiwan, to expand TSMC’s growing facilities in Phoenix, Arizona. Nvidia will also partner with Chinese technology giant Foxconn (maker of iPhones) and Taiwanese company Wistron to build supercomputer manufacturing plants in Texas. Reshoring the semiconductor chip industry back to the US was a major goal of the Biden administration, which passed the CHIPS act that supported the TSMC move to Phoenix, and the Trump administration’s threats of tariffs have lent urgency to this effort.

Nvidia to invest $500 billion to bring its chipmaking back to the US.

Google announces A2A, Agent-to-Agent coordination protocol

As we have been reporting since the end of 2024, when the trend became clear, semi-autonomous Agents are widely accepted as the Next Big Thing in AI. These Agents will accept instructions from the user on the desired end goal, use reasoning to develop a plan, then employ the appropriate tools to accomplish it. One of the “tools” that an Agent might employ is another Agent, perhaps one previously specially trained to perform some subtask in the first Agent’s plan. This will require cooperation among Agents, and as yet, no one knows exactly how best to do that. Google is trying to make all this easier with their new A2A Agent-to-Agent protocol. The protocol is open source, free for anyone to use, and can be used by different developers using different AI frameworks. Google’s A2A is backed by over 50 major companies, including Salesforce, SAP, and PayPal. Google is clearly trying to build an ecosystem that facilitates coordination and collaboration among diverse Agents, increasing demand for Google’s AI models (which they hope to turn into cash cows of the future, rather than the colossally unprofitable cash burners of the present.)

Google announces A2A, Agent-to-Agent coordination protocol

OpenAI says ChatGPT can remember everything from your old chats

One of the annoying foibles of AI chatbots is that they only remember their immediate context. Your previous chat sessions are completely unknown to the standard chatbot once you initiate a new one. OpenAI claims to have fixed that, by giving ChatGPT the ability to reference all of your past chats in any new session. Early users of this new feature report that it does indeed work, after a fashion, but like most other chatbots, the new memory-enhanced ChatGPT loses precision as the context gets larger. So, if you have a lot of old chats, ChatGPT will “sorta” remember them, but it certainly doesn’t exhibit perfect recall. At least not yet.

OpenAI says ChatGPT can remember everything from your old chats

High school student develops AI that discovers 1.5 million space objects

18-year-old high school senior Matteo Paz has won a $250,000 science prize and had a scientific paper accepted for publication for his work using AI to detect 1.5 million previously unknown variable objects in space. Paz, a math whiz who took AP Calculus in the 8th grade, attended a summer science camp at Caltech as a high school freshman, and let his teachers there know that he wanted to do more than play at science - he wanted to participate in meaningful scientific work. Intrigued by his ambition, faculty at Caltech mentored him, and pointed him at a large database that they had been compiling by hand of “variable objects” in space - objects that change brightness due to inherent pulsations, or by being regularly eclipsed by another nearby object. They expected that, like any other teen intern, Paz would slog his way manually through the database to find perhaps a handful of potentially important objects. Paz used his math chops to forge a different path - over the next 2 summers he designed and implemented an AI system that automatically scoured the entire database, eventually finding a staggering 1.5 million previously unknown objects. Based on this achievement, Paz won first place and a $250,000 award in the prestigious Regeneron Science Talent Search, and had a paper accepted for publication in The Astronomical Journal, of which he is the sole author. Still in high school, Paz is now a Caltech employee, doing further work on the database he has mined so successfully.

High school student develops AI that discovers 1.5 million space objects

NATO awards fast track contract to Palantir for AI defense tech

NATO has signed a contract with defense-tech leader Palantir for a customized version of its Maven AI-enabled warfighting command system. The aim is to provide an integrated system for synthesizing intelligence, selecting targets, enhancing battlespace awareness, improving planning, and accelerating decision-making. The contract is said to have been finalized with unprecedented speed, only 6 months from setting requirements to signing the documents. Initial use of the system is expected within 30 days.

NATO awards fast-track contract to Palantir for AI-enabled defense tech

Google AI talks to dolphins

Google’s AI is primarily known for its big, leading-edge models. However, the same Google AI research team that won the Nobel Prize for deciphering protein folding for drug development has now produced a tiny but capable model that can run on your phone. And it speaks to dolphins. Called DolphinGemma, it uses one of Google’s small, open source Gemma models that was specially trained on dolphin vocalizations gathered by the Wild Dolphin Project (WDP), a nonprofit organization that studies the behaviors of Atlantic spotted dolphins. The DolphinGemma AI model, running on a smartphone, can produce synthetic dolphin vocalizations, and listen to dolphin sounds for a matching “reply.” This should help the WDP continue its work to unravel the meaning of dolphin sounds.

Google AI talks to dolphins

Forbes AI 50 for 2025

Forbes business magazine has published its seventh annual AI50 selections, a list of what they deem as the 50 most promising privately-held AI companies in the world. The list is somewhat eclectic, and how “promising” some of the entries are may be a matter of debate, but it is still an interesting list of who’s seen as hot in the AI space at the present moment. The list is alphabetical, so Forbes is making no claim to any definitive ranking.

Forbes list of the 50 hottest AI companies

Boston Dynamics robot dog Spot reactivates radioactive crane

Leading robotics company Boston Dynamics lent its robotic dog, known as Spot, to an effort to reactivate a crane in a nuclear power research site, a maneuver deemed too hazardous for humans. The Dounreay nuclear research site in Scotland was established in the 1950s, and is now engaged in a decommisioning effort that is expected to take decades, due to the high levels of radioactivity contained on the site. Boston Dynamics was asked to assist with its robot for a recent phase in this decommissioning process, and after a week of intensive training on a mock switchboard, Spot proved its worth by successfully using a pole to flip a switch to power up the long-dormant crane. Besides the value of this immediate success, this project points the way to using robots rather than humans in similar hazardous environments.

Spot the robot dog successfully helps decommission a nuclear site

DoorDash partners with Coco for robotic order delivery

On-demand food delivery service company DoorDash is partnering with Coco Robotics to roll out robot order delivery in Los Angeles and Chicago, after a successful pilot in Helsinki. DoorDash’s head of automation and robotics Harrison Shih was quoted saying “Not every delivery needs a 2-ton car just to deliver 2 chicken sandwiches.” DoorDash believes that robotic order delivery will allow it to offer more eco-friendly and more affordable delivery services in select urban neighborhoods.

DoorDash partners with Coco for robotic order delivery

+

How AI can make medicine more human again

Journalist Kate Pickert has written a moving essay in Bloomberg about her experience as a breast cancer patient 10 years ago. Despite having a caring, competent, and generally responsive physician, she found the inevitable delays and frustrations of the medical bureaucracy, and absences of her physician due to vacations or other commitments, to be agonizing at a time when she felt her life was in peril. Looking back, she wonders how much better it might have been if a tool like ChatGPT had existed - available 24/7, always helpful, explaining arcane medical jargon, in a tone of compassion. She notes that most arguments for AI in medicine have focused on efficiency. She makes the case that by lifting the burden of administrative tasks from physicians, they can be freed for what patients need most: attention, care, and an authentic human connection.

How AI can make medicine more human again

Dr. Oz touts AI medicine in first Medicare all-staff meeting

Pop culture celebrity Dr. Mehmet Oz, now head of the Centers for Medicare and Medicaid Services, with a $1.5 trillion annual budget, reportedly spent most of his recent all-staff meeting talking up healthy habits of diet and exercise, and promoting the use of AI agents as frontline health workers that could treat patients at a fraction of the cost of licensed and trained medical personnel. One employee participant was quoted saying “I’m not sure he knows what we do here.”