The DRIVE Series - June 2026
Discoveries, Resources, Ideas, Values, Experiments
Welcome to the DRIVE series, complete with a gimmicky name for my pleasure. At the end of every month, I’ll provide some combination of discoveries, resources, ideas, values, or experiments involving AI, technology, and healthcare. Let’s go!
This month, I’m trying out a new format to put the focus more on ideas. We are covering the early attempts of AI regulation by the government, Anthropic’s concern over recursive self-improvement, and progress (with predictable pushback) on AI in medicine. There’s even a little conspiracy theory news for the enthusiasts.
The Government Lays Down the Law Executive Order
At the beginning of June, President Trump signed an executive order on AI after initially saying that he would not try to regulate AI companies on account of the AI arms race with China. But, likely thanks to Anthropic’s Claude Mythos, the President wants to ensure that government agencies upgrade their American systems for advanced AI within 30 days.
Further, the government will be able to assess and test government systems on frontier models. Interestingly, this is “a voluntary framework” for AI companies. The executive order makes this clear by saying
“Nothing… shall be construed to authorize the creation of a mandatory government licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.”
Basically, the informal agreement between the AI companies and the government is formal: the industry is essentially still unregulated. This is great for innovation and the tech companies’ bank accounts, but probably unlikely to protect the common citizen very much.
Trying to Get a Job Sucks
AI has super powered the ability for job applicants to apply for jobs. Their resumes are practically perfect, their weaknesses are hidden behind active verbs, and, yet, it all looks the same. Meanwhile, the employers are flooded with applications and are struggling to sift through them so they are also resorting to using AI algorithms to sift through the candidates. Everything is becoming monotonous, less personable, and lost in algorithmic soup. I feel like the best way to combat this is to try to meet with people in person, build a network, and create a portfolio of your work. This might mean direct messaging on LinkedIn, cold emailing, and building a portfolio so people can find you.
We’re All Going to Get Hacked
AI cybersecurity will only get more complicated and high stakes with time. Researchers from the University of Toronto claim to have created a computer “worm” that with AI “capable of targeting any known flaw in the world’s computers and quickly spreading mayhem throughout the internet.”
The difference between a computer virus and a worm is that the former spreads through human intervention while the latter spreads on its own. This computer worm wad created using “open-source” technology AI models, meaning the code is available online for free for anyone to view or modify it. The researchers say that they have created a prototype that can tailor its attack to a computer’s specific defense system.
Now, of course, this is a model tested in a lab. Further, AI systems are not perfect and tend to make mistakes. But this should make us all worried that it is not a matter of if we will get hacked one day but when.
For example, even mega-billionaire, venture-capitalist, co-founder of PayPal and Palantir, had his secret, invite-only group called Dialog “hacked,” exposing personal details of members and past participants. A Wired investigation found that the actual cause may have been a mismanaged website, but Dialog claims that it was hacked. Either way, whether due to nefarious reasons or technological mismanagement, if one of the most well-connected and well-resourced technologists is vulnerable, so are any of us.
The Latest on AI Recursive Self-Improvement
This month, Anthropic released a report discussing the concept of “recursive self-improvement,” or the ability of AI to autonomously design and develop without human intervention. In other words, AI that can create other AI that is better than itself.
They make the case that Claude is writing more than 80% of code in Anthropic’s code base, that the quality of the code is improving, that it can test other people’s experiments, and it can propose its own experiments.
Their post makes the caveat that recursive self-improvement has not happened yet and is not inevitable. And while they are calling for a global pause, they aren’t slowing down on their own work because no one is likely to heed this warning anyway. All this comes while they are also planning for a trillion dollar IPO. So who knows what this actually means? Maybe it is an altruistic way to actually get the world to think about how their AI systems might affect us or it is a way to drum up consumer goodwill and investor excitement before the company goes public.
More About the Medical Billing AI Arms Race
This fascinating essay discusses our convoluted medical system in the US and how hospitals and physician groups are locked in a payment battle. Both sides are trying to use AI to improve their position. Hospitals and payment groups are using “revenue cycle management” systems or AI that fights denied claims, anticipates potential refusals, and optimizes billing and coding. Insurance companies are using “program integrity” AI to detect upcoding, deny reimbursements without human review, and flag claims that seem suspicious or also written by AI. Unfortunately, this arms race seems to forget that there is a suffering patient stuck in the middle.
The End of Back Office Jobs or Bullshit Jobs?
This essay makes the case that AI is likely to put at risk millions of “back office,” white-collar jobs. These are administrative positions that include human resources, payroll clerks, bookkeepers, customer service representatives, and other administrative jobs that are not necessarily customer facing. Some of these jobs require college degrees and they are held primarily by women. The concern is that if these positions are replace or eliminated by AI, there will be a large portion of workers who will be unable to climb the career ladder because a rung will have been removed.
I find myself conflicted. Some of the major problems with modern work is administrative bloat in our corporations and healthcare systems. Entire books have been written about the idea of “bullshit jobs,” or jobs that are meaningless, psychologically destructive, and would not be missed if they disappeared. On the other hand, these jobs are held by people who depend on a paycheck. Society would be better off if these jobs were phased out gradually rather than suddenly. Hopefully the AI revolution actually spurs the development of new jobs rather than just erase the existing ones.
Why Prevent For Cheap When You Can Cure for Billions?
Here’s a great example of how our country’s healthcare industrial complex incentives are totally misaligned with reality. Humans have already created very cheap vaccine that can reduce (practically eliminate) the chances of contracting a terrible, preventable measles virus. Yet because of misinformation and disinformation, people are hesitant or refuse to take the vaccine.
So now, biotechnology companies, encouraged by Health Secretary Robert F. Kennedy Jr., want to create and test antibody treatments for measles, because they expect the incidence of measles to rise since people are not being vaccinated. Biotech companies are seeing “a potential multi-billion dollar market opportunity.” Infusions are expected “to cost roughly $2,500.”
In other words, we are ignoring the idea that an ounce of prevention is worth a pound of cure. Companies will take advantage of people’s lack of education, mistrust in medicine, and our current government’s tendency to mislead its citizens in order to make billions of dollars from our suffering. Fun times!
Move Over SEO. GEO is Here
Anyone who has every created a website and tried to get views is aware of the concept of search engine optimization (SEO), which is basically using certain key words or blogging strategies in order to get favored by Google’s search rankings. Since chatbots are now summarizing the internet for users so that search rankings matter less, a new practice called “generative-engine optimization” or GEO is being used. Some tricks include using self-promotional listicles, Reddit posts, and AI recommendation poisoning, which is embedding hidden instructions in “Summarize with AI” buttons. The colloquial term for GEO is sloptimization.
The Future of Electrocardiogram Analysis
A deep learning model trained on over 1 million heart rhythm and imaging records, called EchoNext, was used in a clinical trial at New York Presbyterian-Columbia University Irving Medical Center and detected evidence of severe heart damage in a patient’s electrocardiogram. The patient was called back to the hospital and it was discovered that his heart had an ejection fraction of 10%. Ultimately he ended up with a heart transplant.
EchoNext reads an ECG in less than 10 minutes and analyzes 500,000 ECGs per year. The company, Pathway Labs, was founded to market it. According to the NY Time article, it will be available for free to any doctor who uses the medical chatbot Open Evidence and submits a patient’s ECG.
Conspiracy Theory of the Month
For the conspiracy theorists, STAT news reports that Eli Lilly’s new triple-receptor agonist, retatrutide, was recently approved by the company and by the FDA for a single 79 year-old patient under their “compassionate use” program. Do you happen to know any well-connected 79 year-olds?
China is Catching Up
A Chinese start up called Z.ai recently released a powerful AI model, GLM 5.2, that is inexpensive and open source, which means anyone can use and modify it for free. While some people are concerned about using models developed in China, and the Commerce Department placed Z.ai on their blacklist, some companies can still use the open source system without fear of data being sent back to China if they are careful how they set up their AI infrastructure.
Meanwhile, a massive computing system in Shenzen, China, called LineShine was crowned the world’s fastest. What makes it special is that it uses standard microprocessors and not special graphics processing units (GPUs) that are normally used by supercomputers.
Early Results from the AI Prescription Renewal Partnership Between Doctronic and Utah
I previously wrote about the company Doctronic and how they partnered with the Utah Department of Commerce to introduce an AI system that allows patients to renew prescriptions for non-controlled medications. If the chatbot detects a suspicious request, requires further lab tests, or is uncertain, it can escalate to a doctor. According to Doctronic’s results, the AI system approved 72% of the renewal requests. 91% of reviewed results were approved by a doctor. 9% of cases that required more information were reviewed by a second doctor. 3% of AI approvals were rejected by both physicians.
Understandably and predictably, doctors are upset. Most of the state’s licensing board and the Utah Medical Association is against it. Objections to the program are that tool has not been fully evaluated, there are real risks to patients that are underappreciated, and that the question of liability has not been fully elucidated. Nonetheless, I don’t see this train stopping anytime soon. These AI systems will only get better with time.
AI Companies and Doom Trolling
The computer science professor and author, Cal Newport, wrote a great essay about the phenomenon of “doom trolling,” which is when companies speak of all the problems their technology will cause while saying there is nothing they can do to stop it. He says that this occurs for two reasons.
First is that they actually believe it. If that’s true, then ethically they should stop what they are doing. Second, they do not believe it but they are stressing everyone out because they want to increase their company’s importance or they want to compete for engineering talent or try to gain regulatory permission to continue before any other start ups to impede the latter’s growth.
Doom trolling is a departure from the typical, unbridled Silicon Valley optimism which is what makes it so compelling. But it is exhausting to hear from companies about how their technology will lead to widespread unemployment or societal destruction, especially if it is unlikely to pass. It could be one reason why Americans have such a negative view of AI, while citizens in other countries have a more positive view.
What’s the deal with AI Slop?
This really entertaining and informative article sheds light on the AI slop videos that we are seeing all over the internet from Fruit Love Island to characters like Tung Tung Tung Amir. Apparently, people who have figured out how to create AI slop and manipulate attention are making real money blurring the lines between advertising and entertainment. Some call it art, while others call it mindlessness. The philosophy of “slop” is summed up perfectly in this piece:
Once, slop was just what was fed to pigs: scraps of unwanted food and refuse. It’s not that pigs wouldn’t have enjoyed a salmon fillet; it’s that they were also content to eat old lettuce and bread, so throwing barely edible things together in a trough was the cheapest and easiest way to fatten them up. To understand modern slop, you have to think of humans as consuming content in the same way that pigs consume food. The goal of pig slop is to maximize nutrient intake while minimizing cost; the goal of A.I. slop is to maximize time spent consuming content while minimizing cost. There’s comedy slop, literary slop, art slop, niche slop, slop for kids, political slop — but the substance of slop always matters less than the fact that you’re looking at it.