Oprah hosts “AI and the Future of Us” special on ABC September 12 at 8 PM

Hang it up, naysayers - once Oprah Winfrey hosts a special on something, it is officially A Very Big Deal. Perhaps only a Presidential address from the Oval Office is equivalent in potential for one individual to alter the national conversation on a topic. Oprah apparently plans to interview a variety of tech and media figures, including former Microsoft CEO Bill Gates and current OpenAI CEO Sam Altman. ABC portentously dubs discussions of the future role of AI as “one of the most important global conversations of the 21st Century.” Bombastic, but probably true. Oprah’s ability to speak to the general public in understandable, relatable terms is likely to improve popular understanding of the issues related to this global transition to much smarter machines.

Oprah hosts “AI and the Future of Us” on ABC September 12

Anthropic and OpenAI sign with the US NIST

Two of the major AI-centered companies, Anthropic and OpenAI, have signed an agreement to collaborate with the federal National Institute of Standards and Technology (NIST) on AI safety research, testing, and evaluation. NIST is the US government’s primary technical standards-setting organization, and in recent years has been heavily involved in setting standards for cybersecurity. NIST will be given early (pre-release) access to any new model originating in either of the companies, so that it can test it for safety and reliability. Anthropic and OpenAI are hoping that voluntarily consenting to this federal oversight will blunt efforts to pass restrictive AI laws in Congress and the states. Not so coincidentally, the agreements were signed just as the California legislature passed a sweeping AI regulatory law. No doubt the two companies are hoping that this federal oversight agreement will give California Governor Gavin Newsom enough political cover to veto that bill.

Anthropic and OpenAI sign with the NIST

Yale commits $150 million to AI over the next 5 years

Lots of folks are talking a big AI game. Yale is putting its money where its mouth is. Last week the university announced that it is committing $150 million over the next 5 years to become a world leader in AI. The funds will go to develop and deploy secure AI tools for the entire Yale community (students can cheat on their homework with approved chatbots), create more than 20 new AI faculty positions, distribute seed grants to departments to help them adapt their curriculum to better incorporate AI, encourage cross-departmental collaboration, and greatly expand the computing infrastructure devoted to AI. Although $150 million over 5 years is a serious amount of money, it is only one half of one percent (0.5%) of Yale’s $6 billion annual budget over that time period.

Yale commits $150 million to AI over the next 5 years

Meta reports 350 million downloads of Llama AI model

Mark Zuckerberg is a jerk, but he’s no fool. Zuck became a hero of open-source AI mostly by accident, when the parameters of Meta’s Llama AI model were leaked, without Meta’s consent. Zuck promptly made lemonade out of lemons and embraced open source because 1) he is making gobs of money from his social media properties such as Facebook and Instagram, and doesn’t need AI revenue, and 2) AI startups Anthropic and OpenAI do need AI revenue, and having an open-source alternative constrains how much they can charge, thus weakening them as potential future competitors. Regardless of how he got there, Zuck has leaned into his open-source leadership, developing and releasing ever more powerful Llama models. This has made Llama the most popular open-source AI model available, with 350 million downloads over the past 2 years. Even large enterprises like Accenture, Goldman Sachs, and Zoom use Llama in their AI projects, liking the control and security they gain by running their own model on their own hardware.

Meta reports 350 million downloads of open-source Llama AI model

AI “Insect Eavesdropper” uses crop vibrations to identify pests

Hungry, hungry caterpillars (and other destructive agricultural crop pests) cause an estimated $70 billion worth of damage to crops worldwide. Conventional control methods such as spraying fields with pesticides are currently not targeted enough to have optimum efficacy with acceptable environmental impact. Now researchers at the University of Wisconsin have developed a system that uses contact microphones attached to the crop plant to continuously monitor the vibrations caused by chewing or boring insects. They have trained an AI system to detect the vibrations, identify the species of pest causing them, and count the number of each pest on the plant. Analyzing this data allows farmers to more precisely target pests in time and space, thereby saving money, saving the crop, and helping save the planet.

AI “Insect Eavesdropper” analyzes vibrations to identify crop pests

AI-powered bodycam transcriber automates police reports

Like almost everybody else, police officers hate doing paperwork, and tend to be bad at it. Yet filing crime and arrest reports is an integral part of their job. Now Axon, the company that makes tasers and other police technologies, has developed a system that transcribes interactions from bodycam recordings, and summarizes the information into a standard report format, saving a police officer potentially hours per week of documentation effort. This is another of the AI-powered transcribe-and-summarize products being released for business meeting minutes, physician visit notes, and similar use cases.

AI-powered bodycam transcriber automates police reports

WaPo analyzes what people actually use AI for

The Washington Post has run a story analyzing a research database of over 200,000 interactions between humans and an AI chatbot, aiming to determine how people are actually using AI these days. The top uses of AI were 1) Creative writing and role-play: 21% of requests were for fan fiction, poems, jokes, or role-play. 2) Homework help: 18%. This is probably a lower percentage than conventional wisdom would have predicted. 3) Requests for information: 17%. Due to the problem of hallucination (chatbots sometimes make things up), this is not currently the best use of AI. 4) Work/business: 15%. Another use that is likely overestimated by conventional wisdom.

Although coming in 5th with only 7% of requests, help with generating computer code is likely to increase rapidly in light of recent improvements to both the power and the ease of use of AI coding assistants.

WaPo analyzes what people actually use AI for

UK private school launches first AI teacherless classroom

A posh private school in the UK, where students are prepared for the nationwide exams that determine whether, and where, they can go to university, has now launched a teacherless classroom, where students learn from AI tutors. The AI system continually assesses each individual student’s progress, and will focus on tutoring the student in their weak subjects, so that each student gets an individualized curriculum. There will be adults in the classroom, called “learning coaches” who monitor behavior and teach topics that are more difficult for AI, such as art, sex education, and public speaking. For all this, the students are paying around $35,000 annually.

UK school launches first teacherless classroom

Study shows AI can help less-experienced clinicians “level up”

A large, randomized trial to assess the clinical impact of AI-enhanced cardiac ultrasound interpretation found that expert clinicians did not improve, but that less-experienced clinicians were able to “level up” and equal the performance of more experienced physicians.

Study shows AI can help less-experienced clinicians “level up”

AI helps design highly effective novel CRISPR gene editor

CRISPER gene editing technologies have immense promise, as demonstrated in the recent revolutionary FDA approval of a cure for sickle cell disease. However, CRISPR proteins, taken from bacteria, can be imprecise when used in humans. Now researchers at startup Profluent Bio, which calls itself an “AI-first protein design company” have used AI to design a better-performing CRISPR protein molecule which they call OpenCRISPR-1. Profluent asserts that this achievement not only “expands the CRISPR toolbox,” but also opens the way for development of gene editors tailored to specific applications such as agriculture or medicine.

AI helps design highly effective novel CRISPR editor

UK develops AI to identify patients with higher heart risk

The University of Leeds has trained an AI model on the medical records of over 2 million patients in the UK, developing a system to detect patients with a high risk of developing cardiovascular complications. Having validated the detection aspect of the system, it is now being used in a pilot program where patients identified as high risk are given more intense monitoring and intervention, with the hope of avoiding complications and ultimately saving lives.

UK develops AI to identify patients with higher heart risk

Samsung acquires French AI ultrasound startup Sonio

Electronics giant Samsung has closed a deal to acquire French AI ultrasound startup Sonio. Founded just 4 years ago, Sonio’s first product was an AI system that helps physicians produce ultrasound assessments faster and more accurately. They have recently received FDA approval for another product that uses AI to improve ultrasound images in real time. France is currently a hotbed of AI innovation, and the EU’s best hope for avoiding a US-dominated AI future.

Samsung acquires French AI ultrasound startup Sonio