Stanford posts Top 10 AI trends of 2024

Stanford University’s massive (502 page) annual analysis of the past year’s AI events and trends is out, and here are 3 of the top 10 takeaways:

#1 AI beats humans on English understanding, but lags in reasoning, planning, and math.

#7 AI makes workers more productive and produces higher quality work.

#8 AI is accelerating scientific discovery.

Although a lengthy tome, the report is well-organized and chock-full of insights and data, so even idle browsing is rewarded. (Believe me, I tested this.)

Stanford’s Top 10 AI trends of 2024

Microsoft invests $1.5 billion in UAE-based AI firm

Microsoft has made another bold move in its quest to become the dominant AI company in the world, teaming up with the US government to force the United Arab Emirates into breaking ties with China and bringing its AI under US supervision. Ostensibly, Microsoft made a huge $1.5 billion investment for a minority stake in UAE-based AI company G42, Microsoft’s third largest investment in AI ever. (#1 was OpenAI, and #2 was Cruise, the autonomous vehicle company.) The Middle East oil kingdoms have been pouring petro-dollars into developing their own AI (the new oil), and some kingdoms been playing footsie with China, using Chinese chips and whatnot. Reportedly, the US government made a “If you ain’t with us, you’re agin’ us” ultimatum to the UAE, and they caved. The US appears to view AI as a strategic weapon similar to nuclear weapons, and non-proliferation is the order of the day. Microsoft is installing its President (#2 in the company) Brad Smith on the board of G42 to keep a close eye on its integration with the mother ship. With the US military behind it, what can’t Microsoft do?

Microsoft invests $1.5 billion in UAE-based AI firm

Meta drops AI into Whatsapp, Messenger, and Instagram

Meta/Facebook has earned lots of geek love for open-sourcing their highly capable Llama AI models. But how does Meta plan to make money from its AI? We still don’t know (maybe because they don’t either), but Meta has begun enhancing several of its existing apps with AI features. Whatsapp and Messenger are getting AI chatbots that users can include in their conversations, and Instagram is adding AI-powered filters and ad services. These AI features seem exploratory in nature, introduced with little notice or fanfare. Meta seems in search of traction, with no clear idea of where it might lie.

Meta drops AI chatbots into Whatsapp, Messenger, and Insta

Adobe adds video tools from OpenAI, Pika, Runway

Adobe has been an early adopter of AI among large application companies, stuffing AI goodies into its flagship Photoshop app. Now it is giving an AI makeover to its Premier Pro video app, including features from OpenAI and Runway to help develop B-roll supplementary footage, and from Pika to extend the end of a shot. AI video generation is advancing at supersonic speeds, so Adobe needs to get on the AI bus or get crushed under its wheels.

Adobe adds video tools from OpenAI, Pika, Runway

Udio challenges Suno as the “ChatGPT of music”

Suno showed us that we could generate an entire song or musical piece with a simple prompt, as easy as asking ChatGPT to help us cheat on our homework. Now upstart text-to-video site Udio has upped the stakes, giving another outlet for would-be composers who have no actual talent (like moi.) Udio was founded by 4 former stars from Google’s DeepMind AI research arm, and is backed by top VC firms such as a16z, as well as well as angel investors like musical icon will.i.am.

Try it here for free: Udio.com

Udio challenges Suno as the ChatGPT of music

”Udio is my new obsession” a review by PC Magazine

Texas will use AI to grade student academic readiness tests

Texas’ annual state-mandated achievement tests for all children in public schools are high-stakes exams. A student’s failure to achieve an adequate score can result denial of ability to progress to the next grade, or to graduate from high school. And a low aggregate score for the students of a school may cause that school to be taken over by the state, or closed. The exams include a number of open-ended questions, to which the students write out replies. These questions have traditionally been scored by thousands of seasonal temp workers. Now the Texas Education Agency has decided to score these questions with AI, projected to save $15-20 million per year. Fast-tracking wholesale replacement of humans by automation in a system already plagued with allegations of inequities - what could possibly go wrong?

Texas will use AI to grade student academic readiness tests

OpenAI regains the crown with improved GPT-4 Turbo

For one brief, shining moment, Anthropic’s new flagship model Claude 3 Opus was judged to be ever so slightly better on standard benchmarks than OpenAI’s GPT-4, long the acknowledged category leader. Sic transit gloria. Less than a month later, OpenAI has released an upgraded version of its GPT-4 Turbo edition, and now again sits atop the Hugging Face leaderboard. Depending on your preferences, this is either Return of the King, or The Empire Strikes Back. GPT-4 has been out for over a year, an eternity in AI-time. Suspense is building for the release of GPT-5. Will it be another Great Leap Forward, or just another incremental upgrade? The fate of several companies, and many, many billions of dollars, turns on the answer to that question.

OpenAI regains the crown with improved GPT-4 Turbo

Tech exec predicts “AI girlfriend” will be a $1 billion business

The NSFW side of AI is in the proliferation of sites hawking chatbots customized to act as your AI girlfriend. As we have reported previously, interaction with certain types of AI chatbot personas can have positive effects, reducing loneliness, allowing low-stakes practice of social interactions, even preventing suicide. And, in a story in the AI in Medicine section below, researchers have even entered a psychotherapy chatbot into a clinical trial. But commercializing AI girlfriends smacks of the worst kind of exploitation, encouraging addictive attachment to an online persona at the expense of real world relationships. Here’s hoping the constructive, helpful chatbots win out.

Tech exec predicts “AI girlfriend” will be a $1 billion business

AI can estimate cardiovascular risk from Chest Xrays

Researchers at Harvard-affiliated Massachusetts General Hospital have developed an AI system that can assess cardiovascular risk directly from chest radiographs, which is useful in the common case where certain information needed for standard cardiovascular risk calculators is missing for a particular patient.

AI can estimate cardiovascular risk from Chest Xrays

Dartmouth researchers place psychotherapy bot into clinical trial

Dartmouth psychology researchers have been developing Therabot, an AI-powered text-based psychotherapeutic chatbot, for over 5 years. With recent improvements, Therabot has been entered into a formal clinical trial of its therapeutic effects. The hope is that Therabot will broaden access to psychotherapy, which is in chronically short supply, and be accessible at times a human therapist wouldn’t, such as 3:00 AM. Results of the trial should be known later this year.

Dartmouth researchers place psychotherapy bot into clinical trial

AI assists early detection of pancreatic cancer

Pancreatic cancer is so deadly in part because it causes so few symptoms until its late stages. In a recent article in the Lancet, UK researchers describe how they developed an AI model to detect potentially high-risk individuals by analyzing the electronic medical records of an entire population. The model identified factors from diagnoses, lab tests, and other clinical information that were associated with higher risk. Some factors were intuitive, such as age or recent onset of diabetes, but others were a surprise, such as mean corpuscular hemoglobin in red blood cells. Using the identified risk factors, the model was able to select a group of patients with a 30 to 60-fold higher risk of developing pancreatic cancer in the next 12 months. This study is retrospective only, but the hope is to validate the model with prospective trials.

AI assists early detection of AI cancer

AI reduces complications and costs in radiation therapy

Researchers at Duke and UNC developed a machine learning model that identified patients at risk for complications from radiation therapy. Prospective use of the model to increase evaluations of the patient from once weekly to twice weekly was found to significantly decrease complications, and therefore total health care costs.

Ai reduces complications and costs in radiation therapy