Back in 1959, two researchers wrote a paper in Science basically daring the future to build a computer that could out-diagnose a doctor. The future took its sweet time. This week it finally RSVP'd, and the answer was an awkward "yeah, about that." We've also got a coin-sized brain implant that beat Silicon Valley to the shelf, a free gene therapy, science agencies hoarding cash like dragons, and the most relaxing toxicology result of the year. Let's go.
Table of Contents
🤖 AI out-diagnosed the doctors, and the doctors are taking it well
RESEARCH
Picture an emergency room at 3 am. A patient rolls in with a mess of symptoms, a half-finished chart, and lab values that don't agree with each other. Now imagine handing all of that raw, ugly data to a chatbot and asking it to figure out what's wrong. Turns out the chatbot does it better than the humans.
In a new study in Science, Harvard Medical School researchers fed OpenAI's o1 reasoning model 76 real emergency cases from Beth Israel Deaconess and compared its diagnoses to those of two internal-medicine attendings. At initial triage, the AI nailed the right-or-very-close diagnosis 67% of the time, versus 55% and 50% for the doctors. Give it the full picture at admission, and it climbed to 81.6%.
The clever part is that nobody cleaned the inputs. The model got the same chaotic chart a tired resident would, which is exactly the test that earlier benchmarks dodged by using tidy textbook vignettes. This is the paper that retroactively answers the 1959 challenge, 67 years late.
To be fair, the comparators were two general internists, not ER specialists, and the authors are far more nervous than the headlines. One of them worries the result will be grabbed by companies looking to skip the safe, boring parts of medicine. A model can ace the diagnosis and still order a scan that hurts you.
🧠 China shipped a brain implant while everyone else is still in trials
NEWS
While Silicon Valley spends years and several documentaries' worth of drama getting a brain chip to market, China just put one on the shelf. In March, regulators approved the world's first implantable brain-computer interface available outside a clinical trial, a device called NEO from Shanghai-based Neuracle.
Here's the trade-off worth understanding. NEO is a coin-sized wireless implant that sits on the surface of the brain rather than spiking into it, reading signals from outside the cortex and beaming them to a robotic glove. Less invasive than Neuralink's electrode threads, but also less powerful: it picks up fewer signals, so it does one thing well rather than everything okay.
That one thing is grasping. The device is approved for adults aged 18 to 60 with quadriplegia from cervical spinal cord injury, letting them pick up a cup again. It's the Toyota Corolla of brain implants, which is meant as a compliment: less horsepower, but it's actually street legal.
Meanwhile, the Western rivals are still inside their trials, no FDA, CE, or MHRA approval to show for it yet, and Chinese BCI money is pouring in fast. StairMed alone raised 500 million yuan in April in a rare joint bet by Alibaba and Tencent. The catch, of course, is that "less invasive" still means brain surgery, and the trial data is from press summaries, not peer review. Worth watching, not worshipping.
👂 The first gene therapy for deafness is here, and it's free
NEWS & RESEARCH
Roughly 50 American newborns a year are born unable to hear because of a broken copy of a single gene called OTOF. This week, for the first time, there's a one-time fix. The FDA approved Otarmeni, the first-ever gene therapy for genetic hearing loss, from Regeneron.
The mechanism is straightforward enough to explain at a bar. Otoferlin is the protein that lets inner-ear hair cells talk to the auditory nerve. When the OTOF gene is broken, the conversation never happens. The therapy uses a harmless virus to smuggle a working copy of the gene into those hair cells, restoring the line. In the trial cohort behind the approval, 80% of evaluable patients (16 of 20) hit the hearing target by 24 weeks. The earlier peer-reviewed cut, published in NEJM, had restored meaningful hearing in 11 of 12 children, some going from "can't hear a lawn mower" to "can hear a whisper."
The real plot twist is the price. Regeneron is providing the drug for free to clinically eligible patients in the US. Free as in the gene therapy itself; the surgery to deliver it into the cochlea is still on you and your insurer, so read the fine print before celebrating.
One more wrinkle. The FDA cleared this in 61 days using a new fast-track voucher program, whose public comment period was scheduled for the month after the approval went through. Selling the cake before the committee finishes deciding whether the cake is allowed.
💸 Science agencies got the money. They just won't spend it
NEWS
Plot twist nobody saw coming: after the administration proposed gutting research budgets, Congress pushed back and actually handed NIH a $415 million raise for 2026. Happy ending, right? Except the agencies are now sitting on the cash like it's cursed.
A new analysis argues the agencies have the money but can't seem to get it out the door, written by Jordan Dworkin, now at the philanthropy Coefficient Giving. The numbers are rough: by mid-spring, NSF was making new awards around 70% slower than its usual pace, NIH about 50% slower, and the National Cancer Institute a brutal 79% behind. The money is appropriated. It's just not moving.
Why does a slow start matter if the cash is sitting right there? Because NIH runs on single-year appropriations. Whatever isn't spent by September 30 doesn't roll over, it evaporates and goes back to the Treasury. So a quiet spring isn't a rounding error, it's a countdown timer on a bonfire of unused research dollars.
This isn't unprecedented, which is the worrying part. Grant-making stalled badly in 2025, too, before pressure pried it loose late in the year. The optimistic read is that the agencies pull the same late catch-up again. The pessimistic read is a kid who got the bike for Christmas, won't take it out of the box, and the box self-destructs in September.
🪱 Earthworms don't hoard microplastics, which is at least nice for the worms
RESEARCH
Some good news from the soil, with caveats attached. Researchers at the University of Guelph buried earthworms in dirt spiked with thousands of times the real-world concentration of microplastics, then peeked inside with a synchrotron to see if the plastic was building up in their tissue. It wasn't.
The plastic went in one end and came out the other, fast. In their study in Environmental Toxicology and Chemistry, the worms' bioaccumulation factors landed between 0.023 and 0.058, where anything that truly accumulates, like mercury or DDT, scores above 1. Of the thousands of particles the X-rays spotted inside the worms, not a single one had crossed out of the gut into the body. As one researcher put it, the plastics are ingested and then simply pass through.
The mechanism is unglamorous and effective: worms bundle the plastic into their castings (a polite word for worm poop) and egest it, with a half-life under a single day. Their digestive tract, the team says, is selective enough to keep the particles from migrating inward.
Now the cold water. The "so there may be hope for us" angle is the press framing, not a finding in the paper. The study used micron-sized particles, not the tiny nanoplastics most suspected of sneaking into human tissue, and it ignored the chemical additives that can leach out during transit. Worm guts also aren't human guts. Still, the same researcher is refreshingly skeptical of the scarier microplastics-in-the-brain studies, which is the kind of caution this whole field could use more of.
What ties this week together? Speed, mostly, and who has it. A computer out-reasons two doctors in a messy ER. China ships a brain implant while the West is still filling out paperwork. The FDA approves a gene therapy in 61 days flat. And then, in a perfect bit of cosmic editing, US science agencies remind everyone that having the money and using the money are two completely different skills. The worms, meanwhile, are quietly clearing their plastic in under a day, which makes them more efficient than most federal grant cycles.
The throughline isn't "machines win." It's that the bottleneck is almost never the science anymore. It's the regulation, the funding pipes, the question of who gets blamed when the robot is wrong. Which of these surprised you most? Hit reply and tell us, especially if you're a doctor who wants a rematch against o1. And if a friend would enjoy the existential whiplash, forward this their way.
Keep questioning everything (especially the chart),
P.S. We asked an AI to diagnose our newsletter's problems. It said "61 days is plenty of time to proofread." We're choosing not to take that personally.