Sam Altman, CEO of OpenAI, is having a bad week. Chief among his woes: an expose’ in the New Yorker that shows him to be devious, self-dealing, and motivated primarily by a will to power that many autocrats could only dream of. The article is penned by legendary investigative journalist Ronan Farrow, who broke the original Hollywood sexual assault scandal that sparked the “Me Too” movement. (Son of Mia and Woody Allen - or possibly Frank Sinatra, according to Mia - Farrow has been immersed in the seamy secrets of the rich and famous all his life.)

Farrow focuses largely on the 2023 OpenAI Board coup, in which Altman was first fired by the Board for being “not consistently candid”, then made a stunning comeback within days by mobilizing almost the entirety of the employees to threaten to quit if he was not rehired. The Board caved within days, Altman ended up firing them and replacing them with Board members more to his liking. Did the employees support Altman against the Board so unanimously because he was a beloved figure? In a word, no. Altman was in the midst of a private equity sale that would allow employees to sell their shares, making countless instant millionaires. That bonanza would evaporate if the Board won and returned OpenAI to its not-for-profit roots. Employee loyalty to Sam, as his to them, was ultimately transactional.

In probing this dramatic turn of events, Farrow uncovered a repeated pattern in Altman’s life wherein close associates eventually concluded that he was not to be trusted, because he lied, went back on his word, and made unilateral decisions that benefited him personally with money and/or power. Paul Graham, founder of Silicon Valley’s elite startup incubator YCombinator, eventually had to fire Altman as CEO of that organization due to ongoing complaints that Altman was using his position to benefit his own personal investments over those of the firm.

Farrow’s article is titled with a question: “Sam Altman May Control Our Future - Can He Be Trusted?”, and he concludes that, no, he really can’t be. Farrow calls for the public and the government to stop treating Altman as a benevolent boy-wizard, and to put in place effective oversight and accountability measures for OpenAI and all major AI companies.

Sam Altman is “unconstrained by the truth”

Last week US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an emergency meeting with the CEOs of America’s 8 largest banks, to warn them of the danger posed by Anthropic’s superhacker AI model Mythos.

Anthropic has been rolling out its newest model, codenamed “Mythos”, in a very controlled way, due to the startling capabilities the model showed in internal testing after training. The model had been designed and trained to be very good at coding software. In testing, they found that it was so good at coding and analyzing software that they had inadvertently created the world’s first AI superhacker.

The model immediately found thousands of bugs and vulnerabilities in commonly used open source software, some of which had lain undiscovered for decades. The company decided that it would give initial access to just a few dozen organizations with large code bases to defend. This list included Google, Apple, Microsoft, and the Linux Foundation.

Shockwaves from Anthropic’s announcement of the Mythos model’s cyberhacking superpowers have begun to ring alarm bells in financial circles. Last Tuesday, the Treasury Secretary and the Fed Chair jointly called an emergency meeting with the CEOs of the nation’s 8 largest banks, including Bank of America, Citigroup, JP Morgan Chase, Goldman Sachs, and their ilk. These top brass were already in Washington for the day, attending the Financial Services Forum, a trade group that enables them to collude against the public facilitates communication about broad areas of policy. So it was just a quick change of plans for the evening, and the limos were heading for the Treasury building. There the CEOs had their dinner spoiled by hair-raising news of new threats from AI-enabled cyberattacks. No doubt suitably sobered by the news, the CEOs went back to their evenings to drink.

A week later, the British AI Security Institute, a government agency in the UK, confirmed that under testing, Mythos was able to autonomously plan and execute a successful 32-step cyberattack against a simulated corporate network. It appears that Mythos is no myth - its cybersecurity skills are real.

Treasury and the Fed assemble the bankers on a moment’s notice

Anthropic was founded by a breakaway faction of disaffected OpenAI employees who had been lured with promises of high-minded pursuit of the good of humanity, only to find that CEO Sam Altman was casting all of that aside in order to make the company a gigantic commercial success. (See the Top Story above.)

As the AI race heated up, it became (mostly) a three-way race between OpenAI, who had kicked off the whole AI explosion with ChatGPT in November of 2022, Google, who had actually invented the Transformer technology that undergirds all current top AI models, and Anthropic, the new kid with a lot to prove.

Now recent financial estimates show that Anthropic revenues are beginning to outpace those of OpenAI, who had what seemed a commanding head start until just a few months ago. Recent estimates are that Anthropic has a $30 billion annualized revenue which is exponentially increasing, while OpenAI has an annualized revenue of approximately $25 billion which is also growing quickly, but slower than Anthropic. In secondary markets, brokers have more sellers than buyers for shares of OpenAI, while at the same time there are far more would-be buyers of Anthropic than sellers. The spot price for an Anthropic share briefly spiked to a level that would value the whole company at $863 billion, to OpenAI’s $846 billion.

Why the difference? Anthropic has been laser-focused on making valuable tools for business, and therefore has gained an ever-growing customer base among large enterprises that can cut big checks. OpenAI has been trying to build a consumer app where most of its users pay nothing. Now OpenAI is scrambling to get back in the enterprise software game that Anthropic is dominating. So far, OpenAI’s one big win in that sector was grabbing the Pentagon contract when Defense Secretary Pete Hegseth booted Anthropic for refusing to allow the DOD to use its software for mass surveillance of the American people, or to power autonomous killer robots.

Anthropic surpasses OpenAI in revenue and valuation

I have to admit, I had a gnawing sense of dread during the whole Artemis mission. It had been over 50 (!) years since America had had a successful moon mission, and the intervening years had not been kind to either the space program or to the nation. I kept waiting for what seemed the all-too-likely disaster, bigotry, incompetence, or hatefulness that seem to be our daily diet these days.

And none of it happened.

To my surprise and delight the Artemis mission just worked, proving that America is still capable of great things. Everyone involved was clearly competent and good at their jobs. There was no billionaire riding his personal rocket as an ego trip, but a public project, publicly funded, for the benefit of all. The crew included a woman and a black man in positions of leadership, and no one made racist or misogynistic remarks. New scientific advances were made, because science is real and should be funded.

Because of all this, for a brief and shining instant, Americans of every persuasion could feel a moment of unity and hope, holding at bay the ongoing dumpster fire of the current world situation.

And for that I, and many other Americans, perhaps even a majority, are grateful.

Apple’s AI woes have left Siri still dumb as a rock. But now there is a plan for an AI-powered Smart Siri, and it is scheduled to be released with the new IOS 27 operating system this September.

The new Smart Siri will be more conversational, allowing for back-and-forth interactions over time. It will be able to execute multi-step tasks from a single command (such as “Find the nearest coffee shop, add it to my calendar, and text Mike the address.”) And it will be able to read on-screen text, summarize web search results, and edit photos directly within Apple’s native apps.

Since Smart Siri will eventually be pushed out to all of Apple’s 1.5 billion users, it may well become the most widely used interface to AI in the world.

Smart Siri is coming this fall

Uber has committed over $10 billion to buying autonomous vehicles, as well as stakes in their manufacturers. This represents a shift in strategy from its asset-light “gig economy” model of the past, where the contract drivers owned the cars they drove. The shift is being forced on Uber due to the success and expansion of Waymo, the Google-backed robotaxi company that is now operating in 11 US cities, as well as the threat of Elon Musk’s Cybercab, which he promises to sell for less than $30,000 by 2027. (But, you know about Elon and his promises…)

Uber is positioning itself as a marketplace for multiple driverless vehicle manufacturers, so that it doesn’t have to commit to just one. This is a possible winning strategy if the driverless vehicle market is fragmented, so that no one manufacturer has the scale to make developing or purchasing ride-hailing software an economically rational decision. However, in an age of AI-powered software development, that moat might not be as deep as Uber thinks it is.

Uber commits $10 billion to robotaxis

A scrappy UK startup with a Ukraine connection trounced a crowded field of 27 military drone manufacturers, mostly from the US, who were pitted against one another in battlefield competition for a piece of a rich $140 million contract to supply attack drones to the US military.

The UK firm Skycutter, has a relationship with Ukraine drone manufacturer Skyfall, and the real-world experience in drone warfare proved decisive. In a series of 3 increasingly difficult battlefield scenarios, Skycutter’s Shrike 10 Fiber drone scored a near-perfect 99.3 out of 100 possible points. The nearest competitor was more than 12 points lower. Skycutter will therefore get an order for approximately 2,500 drones at $5,000 apiece, for a total order of $12.5 million.

The Skycutter drone is guided through an attached optical fiber, which can be up to 12 miles long, making it immune to radio jamming and GPS spoofing.

The Brits will feast tonight, in celebration of their victory. The US firms are eating humble pie.

UK drone startup Skycutter trounces US firms in Pentagon test

Researchers from the University of Geneva have developed an AI-powered stool test that can detect colorectal cancer from a simple stool test. Using a machine-learning model of how the gut microbiome changes in the presence of colon cancer, the test is 90% accurate, which compares favorably with colonoscopy’s 94% accuracy.

Colorectal cancer is the second-leading cause of cancer-related deaths worldwide, and colonoscopy is the gold-standard way of detecting and treating it. However, the expense, logistical complexity, and the substantial “ick factor” of having a tube inserted into one’s rectum all contribute to making colonoscopy much less used than would be optimal for cancer prevention.

Having a simple stool test compete in accuracy with colonoscopy for detecting colorectal cancer could vastly increase the number of individuals appropriately screened, and make a meaningful dent in colorectal cancer deaths.

AI stool test 90% accurate for colorectal cancer

Speaking at a conference at the University of Maine, Michael F. Chiang MD, Director of the National Eye Institute at NIH, detailed how artificial intelligence, along with advanced data analytics and other digital tools, can be used to improve and expand health care services in rural areas such as much of Maine.

He pointed out that AI, combined with other digital technologies, can move care beyond traditional clinical settings through telehealth, remote monitoring, and home-based sensors and tools. AI-enhanced diagnostic tools can improve diagnosis of difficult cases in rural areas to the level of communities with many more physicians and specialists. And AI tools that relieve physicians of administrative burdens can free up physician time in rural areas to attend to more clinical tasks.

NIH looking to AI to expand rural health care