Trump posts fake AI-generated endorsement from Taylor Swift

Fears of a tsunami of AI-generated fakes in the current Presidential election have proven to be, if anything, understated. Luckily, so far the biggest miscreants, Elon Musk and Donald Trump, have been bad at it - creating images and deepfake voice clones that are neither believable, or clever, or funny. This past weekend, it was The Donald’s turn to float an AI-powered fake endorsement of his candidacy by pop icon Taylor Swift, including images of crowds of “Swifties for Trump” (see below) and one of Swift herself in a be-sequined Uncle Sam outfit urging fans to vote for the Orange One. Trump no doubt envies Swift’s popularity, her bank account, and her ability to sell out arenas.

Trump posts fake AI-generated endorsement from Taylor Swift

Midjourney releases web-based image editor

Midjourney has long been deemed to be the most capable text-to-image app, but it has suffered from a usability problem. It stayed stubbornly walled off in a server on the somewhat arcane Discord gamer chat network, long after many thought that it should become a more mainstream web application. Partly this was a consequence of its refusal to take venture capital funding, so it has had to bootstrap its growth, including computing infrastructure, from monthly subscription fees. Meanwhile, OpenAI got $13 billion from Microsoft alone. In recent months, however, Midjourney has taken steps to develop a web presence, and this past week it improved and simplified the website’s user interface, and stuffed it with more powerful image editing tools, a chronic weakness of the product.

Midjourney releases web-based image editor

Elon’s Grok-2 AI model is capable, but has no guardrails

Last week Elon upgraded his AI model, named “Grok", and the Grok-2 version performs at a level equivalent to the best competitors, such as OpenAI’s ChatGPT, Google’s Gemini, or Anthropic’s Claude Opus. In line with its owner’s self-styled “free speech absolutism,” however, it differs from its peers by allowing apparently unfettered image generation of political figures, celebrities, and cartoons. Users taking advantage of this leeway have posted all manner of quirky, bizarre, or defamatory images, from the tame “Baroque Obama” shown below, to an image of presidential rivals Harris and Trump holding hands affectionately, with the fake Harris obviously pregnant. Election laws allow quite a bit of license to candidates, but copyright holders such as Disney are likely to take a very dim view of Grok users who create images of their characters engaged in unspeakable activities, as has already occurred.

Elon’s Grok-2 AI model is capable, but has no guardrails

Smuggling AI chips to China has become a major industry

Currently, the US government views US dominance of AI technology as a national security imperative, and China as a major rival. Last October, the Biden administration banned exports of the most capable AI computer chips to China. Now almost a year later, evidence is growing that Chinese firms are routinely circumventing these rules, either by creating an opaque chain of shell corporations in other countries, or by the simple expedient of renting cutting edge servers from Amazon or Google, since cloud services are not included in the ban. Shell corporation purchases as large as $100 million per transaction have been widely reported. The degree to which Nvidia, the world’s preeminent AI chipmaker, is turning a blind eye to such black-market transactions is unclear.

Smuggling AI chips to China has become a major industry

Cheyenne WY mayoral candidate plans to govern by AI bot

Victor Miller, a candidate for Mayor in Cheyenne, Wyoming, claims that he is running as a “meat avatar” for the real power behind the throne, an AI bot based on ChatGPT-4, which Miller has named VIC (for Virtual Integrated Citizen). Although Wyoming law requires candidates to be registered voters, Miller vows that he will delegate all decisions to the AI, which will be able to digest the reams of reports required to run the city every month, and make decisions in a more objective way. Miller sees this as a way to replace politicians with AI, and “reap the peace and prosperity that follows.” It appears that very few of his fellow citizens agreed with his plan - yesterday Miller garnered only 327 votes out of 11,000 cast in the mayoral primary.

Walmart uses AI to 100x productivity

Walmart revealed last week that it is using AI to update its massive product catalog, which contains over 850 million pieces of information. Walmart CEO Doug McMillon told financial analysts that this work would have required 100 times more employee work hours to complete than with the AI-enabled process.

Walmart uses AI to 100x productivity

Pentagon plans drone hellscape to defend Taiwan

Cheap weaponized drones have transformed the fields of battle in Ukraine and the Middle East. Now the Pentagon is mulling extension of the lessons learned in these hot spots to the tinderbox of China-Taiwan relations. Many in the US national security establishment think a Chinese invasion of the nearby island of Taiwan in the next few years as a distinct possibility. Takeover of Taiwan would achieve a long-held Chinese goal of reuniting this breakaway island with the mainland, and at the same time, take over the world’s most advanced AI chip-making facility, TSMC, which is located there. In order to slow any Chinese onslaught against the island enough to allow help to arrive from the US and other allies, the Department of Defense is drawing up plans to flood the 112-mile wide Taiwan Strait with thousands upon thousands of drones in the event of any conflict. Drones are comparatively cheap, expendable, highly mobile, and can inflict outsize damage to ships and other large and expensive war materiel on the opposing side. Although submarines may be likened to deadly killer whales, drones are more like a swarm of piranhas. Contact with either is likely to be lethal.

Pentagon plans a drone hellscape to defend Taiwan

NASCAR crews use AI for the perfect pitstop

NASCAR auto racing has evolved to a game of looking for tiny advantages in every aspect of the race possible. One area that has long resisted precise optimization is the pit stop, where the race car is refueled and tires swapped in a matter of seconds. Now Lenovo has teamed with Richard Childress Racing to use AI to 1) precisely estimate the amount of fuel needed at each pit stop, and 2) analyze the real time video feed of a camera over the pit to devise the optimum deployment of tools, fuel, and personnel.

NASCAR crews use AI for the perfect pit stop

UC Davis uses AI to give ALS patient his voice back

Researchers at UC Davis have developed a new brain-computer interface (BCI) which can translate brain signals into speech with up to 97% accuracy. Researchers implanted sensors into the brain of a man whose speech muscles were paralyzed by the neurodegenerative disease Amyotrophic Lateral Sclerosis (ALS). The sensors were placed in the parts of the brain which control the speech muscles, and by tracking activation sequences of these areas, the researchers were able to deduce which sounds the patient was trying to produce, and therefore which word. This method greatly simplified and speeded up the process of decoding the brain signals into English words, and then produce a text “transcript.” This transcript was then read aloud by an AI voice that had been trained to be a clone of the patient’s former voice, based on recordings. With 30 minutes of training, the patient could “speak” a 50-word vocabulary with 99.6% accuracy. With an additional 1.4 hours of training, the patient mastered a 125,000 word vocabulary, with an accuracy of 90.2%. Subsequent training sessions have increased accuracy to 97%.

UC Davis uses AI to give ALS patient his voice back

NEJM article

Kaiser, Northwestern give AI scribes to thousands of clinicians

AI Scribes are sweeping the nation. Recently Kaiser Permanente began rolling out AI technology to automate documentation of patient visits to 24,000 clinicians at 40 hospitals and 600+ clinics. Kaiser is deploying the system developed by AI startup Abridge, one of the two leading AI Scribe vendors. The other major vendor in the US is Nuance, a Microsoft subsidiary, which is partnering with Chicago-based Northwestern Medicine on a demonstration project to test Nuance’s DAX Copilot AI Scribe system. Initial results of the Northwestern trial showed physician users experience a 17% decrease in note time, a 24% decrease in “pajama time” after-hours work, and an additional 11.3 patients seen per month.

Kaiser rolls out AI Scribe Technology to thousands of clinicians

Northwestern Medicine demonstration project shows positive results

UCLA trains AI to detect rare immune disease

Researchers at UCLA have trained an AI system to flag patients with possible Combined Variable Immunodeficiency Disease (CVID) through detecting subtle patterns in their electronic chart. CVID is rare, and both its causes and its symptoms can vary greatly from one case to another. This often results in delayed diagnosis, sometimes for years. Delayed diagnosis increases the patient’s number of infections and complications, increasing morbidity and total health care costs. Researchers hope that the AI system for detecting CVID will be applicable to other rare and hard-to-diagnose conditions.

UCLA trains AI to detect rare immune disease

AI antibody generator Absci inks deal with Sloan Kettering

Absci, an AI-powered drug design company, has joined with Memorial Sloan Kettering to co-develop up to 6 novel immune therapies for cancer. This collaboration follows similar co-development deals Absci has signed with AstraZeneca and Merck, each of which is expected to produce hundreds of millions of dollars in revenue to Absci.

AI antibody generator inks cancer collaboration with Sloan Kettering