New Post 3-6-24

Top Story

It’s war! Elon sues OpenAI and Sam Altman

Elon Musk, world’s richest man-baby, threw another tantrum this past week and filed suit against OpenAI, as well as CEO Sam Altman and co-founder Greg Brockman, for breaking the charter of the original nonprofit organization that the current for-profit company came out of. The lawsuit is stupid in so many ways, and is sure to lose in court, but it is a helluva publicity stunt for Baby Elon, who appears to believe that there is no such thing as bad press.

He may yet be shown the error of his ways. Realizing that they were in a war for public opinion, Open AI printed a batch of Elon’s emails showing him to be variously 1) greedy, 2) power-mad, 3) hypocritical, and 4) a liar. Keep your popcorn handy - this show may be a doozy.

“Why are these guys getting attention? I want ALL the attention!”

Clash of the Titans

Anthropic releases Claude 3 “family” of AI models

Anthropic (founded by scientists who left OpenAI over AI safety concerns) have released their latest upgrade to their flagship competitor to ChatGPT, Claude. Claude now comes in 3 sizes: small (codenamed Haiku), medium (Sonnet), and large (Opus). Opus is broadly equivalent to OpenAI’s most sophisticated model, GPT-4, as is Google’s Gemini Ultra. These three are currently the leading “foundational” AI models, and each is paywalled behind a $20 per month subscription fee. Each company makes claims to offer the best model, but the differences will in general only be important to heavy commercial users.

ChatGPT can now talk to you

This week, OpenAI rolled out voice response for all ChatGPT users. Now you can have ChatGPT read its responses to your prompts in one of several user-selected voices. This is available on the web, and on both Android and IOS mobile devices.

Amazon buys nuclear-powered data center

AI is an energy hog. By some estimates, AI could triple the share of US energy devoted to data centers by 2030. AI companies, and the cloud providers that service them, are scrambling to secure access to the energy they will need. Amazon took the direct approach, and bought a data center that was already attached to the nation’s 6th largest nuclear power plant, the Susquehanna Steam Electric Station in Salem, Pennsylvania.

With the purchase of the data center, Amazon acquired the right to use up to 960 megawatts of the nuclear plant’s electricity production, approximately 40% of the plant’s 2.5 gigawatt capacity. This is enough electricity to power a small city. Never say that Jeff Bezos thinks small.

“FEED ME!” says Amazon’s new power-guzzling data center.

Fun News

California allows Waymo to expand its robotaxi service

California regulators have given Waymo permission to expand its self-driving robotaxi service to the highways in Los Angeles and the Bay Area. This is a swift reversal of their decision last month to put the expansion on pause after a few high-profile accidents involving the driverless vehicles. The politics of autonomous vehicles in California is tangled, with supporters and critics equally fervent, so despite this decision, Waymo likely faces a bumpy road ahead as it strives to make its service ubiquitous.

When you just can’t face having to talk to one more Uber driver…

More Sora text-to-video eye candy

Sora, OpenAI’s new text-to-video phenomenon, keeps producing viral clips on X/Twitter. One of the latest is this eye-popping fly-through of an imaginary museum. Go ahead, click the link and see,

“I’m actually a real museum. Click the link to see the imaginary fly-through.”

Google’s DeepMind AI lab demos a videogame generator

Legendary AI development lab DeepMind has just demonstrated Genie, a nifty videogame generator, that can take a text prompt or an image - even a sketch - and turn it into a playable videogame. Most of the gameplay looks like a Mario knockoff, but the demo is still impressive. Genie was apparently trained on 300,000 hours of videogame play.

US Army testing AI as battlefield advisors

The US Army is reported to be assessing the battle planning abilities of AI by testing it in the military science fiction videogame Starcraft II.  Researchers at the US Army Research Laboratory are using commercially available AI chatbots to act as battle command assistants, quickly proposing alternative actions and responses. The Army deems its researches “promising”, even though human players are currently beating the bots handily.

AI in Medicine

AI discovers that prostate cancer is more than one disease

UK researchers have used AI to determine that prostate cancer actually has two distinct subtypes. This new understanding of how prostate cancers evolve is hoped to lead to new, more targeted and more effective therapies. The sophisticated analytical techniques used in this study are also deemed likely to help inform our understanding of a wide variety of other cancers.

AI smartphone app diagnoses ear infection

An AI-powered smartphone app developed by researchers at the University of Pittsburgh diagnoses ear infections with an accuracy of 94%, which is better than most physicians. The AI analyzes short videos of the eardrum captured by an otoscope attached to the smartphone. An associated article in JAMA Pediatrics indicates that this tool could meaningfully decrease unnecessary use of antibiotics in children.

AI predicts kidney failure 6 times faster than human experts

Nephrologists and scientists at Sheffield Teaching Hospital in the UK have developed an AI program that assesses kidney volume in patients with Autosomal Dominant Polycystic Kidney Disease, a key measurement in gauging the progression of this disorder.

Currently, clinicians have to perform this measurement manually, by painstakingly tracing the outline of the kidney on slices of an MRI scan one by one. This can take up to an hour per patient. The AI system performs this task in less than a minute.

That's a wrap! More news next week.