OpenAI CEO Sam Altman wants $7 trillion

Well, so do I, and so does everybody! Sam’s a lot more likely to get it, tho. Altman, an inveterate and peripatetic dealmaker, has been pitching to Middle Eastern sovereign wealth funds for up to $7 trillion (approximately 7% of the world economy) to build more factories to make the computer chips that fuel the AI revolution. Sam is nothing if not ambitious. If Altman succeeds even partially in this effort, it will profoundly change the dynamics of the chipmaking sector, and possibly of the entire technology industry.

Here’s a CNBC report on Sam’s $7 trillion quest

For a deeper dive on why Sam thinks he needs the $7 trillion

Google (finally!) releases Gemini Ultra, its next-gen LLM

The sleeping giant stirs. Despite having literally invented the Transformer neural network architecture that underlies all modern AI, Google was caught flatfooted by the wild success of ChatGPT, and has been mostly fumbling the ball ever since. They have kept bragging about how good Gemini was going to be, but kept missing release dates. Now their response to ChatGPT is out, and… it’s pretty good. Not killer, but more or less equal to GPT-4, which is currently the reigning king of the AI hill. Time will tell whether Google can leverage all their other advantages (pre-eminence in search, bazillions of daily users for a suite of apps, torrents of cash flow, etc. etc.) to make a “good enough” AI a winner.

Google finally releases Gemini Ultra, its next-gen LLM

OpenAI gives ChatGPT memory, and soon, agents to do stuff

OpenAI is an app-building, app-shipping machine. It is now working 2 key upgrades to its flagship ChatGPT product address two of the major dissatisfactions users have had with the world’s most popular chatbot.

First, there’s memory. Until now, ChatGPT has not remembered information you give it from one chat to another. Even in a single chat session, if it goes on long enough, ChatGPT begins to lose the plot. This has to do with its underlying architecture, which makes it computationally costly to keep track of too much context. OpenAI has announced that it is now adding user-controlled memory to ChatGPT, so you can tell it what to remember, and what to forget from your chats.

Second, agents. Up until now, ChatGPT was all talk and no action. You, the user, had to do all the messy file uploading, web searching, and assembling the final product of all these efforts. Now, as reliably reported in tech news site The Information, OpenAI is working on add-on capabilities that allow ChatGPT to take over your computer for the duration of the task, and complete the work itself.

ChatGPT gets a user-controlled memory

OpenAI is working on giving ChatGPT agents to automate tasks

Amazon lost $40 billion on Alexa, soon to be replaced by AI

Amazon had high hopes for the voice assistant Alexa. At one time, it is reported that it had a team of thousands working on the project. Over the years, it appears to have lost over $40 billion on this product. With the advent of Generative AI like ChatGPT, most of that effort will turn out to have been wasted, and the new Alexa, or Alexa’s successor, will be powered by AI.

Amazon lost $40 billion on Alexa, soon to be replaced by AI

Your AI “girlfriend” is probably a data-harvesting horror show

Seeking romance through a commercial transaction has always been fraught with peril. Just ask Eliot Spitzer, former Governor of New York, forced to resign for using escorts. Online romance is no different. The fast-burgeoning sector of personal AI chatbots who act as a virtual “girlfriend” or (especially in Asia) “boyfriend” is starting to raise alarms about seedy practices, such as harvesting and selling personal data, and fueling dependency and addiction to the service. This is especially unfortunate, not only because it preys on the lonely and socially isolated, but also because studies show that, done right, these AI “friends” can have positive impacts on users’ socialization and mental health. One study of 1000 users of an early version of virtual friend concluded that it prevented 30 suicides. (link)

Your AI “girlfriend is probably a data-harvesting horror show

Men turn to AI to write their Valentine’s Day love message

Chatbots simulate sincerity and empathy really well. Better than most men. Recognizing this, an increasing number of men are outsourcing their intimate messages (or their profiles on dating apps for the unattached) to ChatGPT or similar apps. Business Wire reports that 45% of men are contemplating using AI to write their Valentine’s Day love message. Women also use AI for intimate messages or profiles on dating apps, but at about half the rate of men.

Men turn to AI to write their Valentine’s Day love message

TikTok releases Boximator image to video app

Social media video powerhouse TikTok has developed a simple image-to-video app, called Boximator. Take any image, draw boxes around the part you want to automate, draw an arrow for the direction you want that object to move, and - voila! - a simple but effective animation will result.

Demo video of Boximator image to video

Pakistan opposition party uses AI to win big

The ruling party in Pakistan used every trick in the book to stay in power, even jailing the opposition party leader on trumped-up charges. The opposition party fought back with AI - and won a landslide victory. First, the opposition party used voice cloning technology to deliver speeches written by the jailed opposition leader on radio. Then, when the ruling party made last minute changes to polling sites, and used random symbols for the opposition party on the ballot (different symbols for different sites), the opposition set up a Facebook page as an AI chatbot. Voters could message the chatbot with their National ID card number and their constituency, and the chatbot would reply with the voter’s polling site and the party’s symbol for that site. Naturally, the victorious opposition leader, still in jail, used AI voice cloning technology to deliver his victory speech.

Pakistan opposition party uses AI to win a landslide victory at the polls

Over half of marketers surveyed say AI tools “integral” to work

Instapage’s State of AI in Marketing report indicates that AI has made deep inroads into marketing departments, with 55% of those surveyed saying AI tools were “integrated pretty deeply into my daily workflow,” and another 16% saying that they “can’t live without them.” AI chatbots excel at plausible-sounding text that’s not necessarily true, so it’s no wonder AI has found a warm welcome in marketing.

Over half of marketers surveyed say AI tools “integral” to their work

Georgia Tech researchers develop blood test for ovarian cancer

Ovarian cancer is usually asymptomatic until it’s too advanced to cure. Researchers from Georgia Tech have used sophisticated machine learning techniques to develop a blood test for the disease. The test was able to distinguish cancer patients from unaffected controls with 93% accuracy. Much work remains to further validate the test and confirm its clinical utility.

Georgia Tech researchers develop blood test for ovarian cancer

An AI-designed drug to be tested on humans, in a world first

UK biotech startup Exscientia used AI to design a drug for obsessive-compulsive disorder. It will soon enter Phase 1 trials in Japan, with further trials planned globally if this one is successful. This will be the first time an AI-designed drug will be tested in humans, according to company sources.

An AI-designed drug to be tested on humans, in a world first

Major health systems experiment with Vision Pro headset

Apple’s new Vision Pro “mixed reality” headset has created a stir with consumers, but several major health systems have seen a number of compelling uses in health care. Here a few examples.

Sharp Healthcare of San Diego has reportedly purchased 30 Vision Pros and has partnered with dominant Electronic Medical Record vendor Epic to explore uses.

Sharp Healthcare partners with Epic for uses of Vision Pro

Cedars-Sinai of Los Angeles has developed a mental health app for the Vision Pro that offers an immersive self-administered therapeutic experience with an AI avatar at a calming virtual site such as a beach.

Cedars-Sinai develops immersive therapeutic experience with AI avatar

Boston Children’s Hospital develops an immersive training app for nurses and other health care workers.

Boston Children’s develops immersive training app

CMS bans AI for denials of coverage by health plans

CMS has notified Medicare Advantage health plans that they cannot use algorithms or artificial intelligence to determine care or deny coverage to members. This comes in the wake of multiple suits by patients against UnitedHealth and Humana for using an AI tool called nH Predict, developed by a subsidiary of United Health. The allegations are that the AI tool routinely recommended denials of claims for post-hospital care after 14 days, despite the benefit being up to 100 days, and that health plan employees were not allowed to deviate from the recommendations of the software.

CMS bans AI for denials of coverage by Medicare Advantage plans