Edward Bukstel

Awesome time with awesome people at @Redox in #Philly. Amazing company with #healthcare APIs. Thank you @ShereeseMayMba and @eileengadsden for meeting at @philamuseum Where’s @techguy ? #HCldr

BIG claim from new MIT + Oxford + Carnegie Mellon and other top labs paper: AI can boost performance at first and then leave people less able to think through problems on their own. Just minutes of AI help can improve scores now while weakening independent problem-solving right after. The interesting part is that the damage is not just lower accuracy. It is lower persistence, which is usually the hidden engine of learning, because skill grows through repeated contact with difficulty, not just exposure to correct answers. That's why a good teacher sometimes withholds help to preserve struggle as part of the lesson, while today’s chatbots are tuned to erase friction on demand. Across 3 experiments in math and reading, about 1.2K people either worked alone or used a GPT-5-based assistant for part of the task. Assisted users finished early questions faster, but after roughly 10 minutes without AI, they solved less, stalled more, and quit sooner. That happens because hard thinking is not only about getting answers; it is also about building the habit of holding a problem in mind, testing steps, and pushing through confusion. The sharpest drop came from people who used the model for direct answers, not from those who used it more like a hint system, which suggests the real issue is not AI exposure itself but replacing effort with completion. The result is not that AI makes people less capable by default, but that answer outsourcing can shrink the mental effort that normally trains skill. ---- Paper Link – arxiv. org/abs/2604.04721 Paper Title: "AI Assistance Reduces Persistence and Hurts Independent Performance"

Facts. Most AI safety people aren’t smart enough to understand what is going on.

On the "law should be public" and the "access-to-justice" fronts, Legaltech News (LTN) has a post on the good work being done by Free Law Project: From the LTN post: "On Thursday [April 16, 2026], Free Law Project (FLP), a nonprofit dedicated to expanding access to primary legal materials, announced a new initiative to scan and digitize case law from commercially published volumes of opinions. "The cases can then be added to CourtListener, the organization’s archive of court data. "FLP executive director Mike Lissner said that FLP has already digitized over 200,000 pages of historical case law out of a total of roughly 2.5 million pages of opinions published since 2018. He expects the work to be completed sometime this fall. Read the complete LTN post here: law.com/legaltechnews/… #legaltechnology #legaltech @FreeLawProject

@anothercohen 😝😝

"Hi, thanks for calling Dr. Johnson's office, please hold"

@ZachAbramowitz I stand corrected! 😝

@ebukstel More like inside voice is LinkedIn, tipsy at a bathhouse voice is X

I gotta be careful about what I say on LinkedIn after first hanging out on X. I just accidentally told the truth over there in a comment on someone's post

@andrewarruda His metaverse office was deleted by his own action

AI.

When a giant actor like Michael Douglas speaks, the entire media landscape trembles. He’s not the kind of actor who posts a story and deletes it an hour later. He doesn’t write vague tweets that can be interpreted in every direction. Michael Douglas stood in front of cameras and stated plainly: the world has completely lost its moral compass. Israel, he said, is on the front line of the struggle. Not a struggle over territory. Not a struggle for power. A struggle over the values of the entire Western world—democracy, freedom, human rights. All the things people love to talk about on Twitter but aren’t willing to fight for. He outright rejected, without hesitation, any attempt to compare the IDF to extremist terrorist organizations. Such a comparison, he said, is an insult to reality. One side fires rockets at civilians and hides behind children. The other side calls civilians before a strike and tells them to leave. Comparing the two isn’t criticism—it’s moral blindness. And then he said the sentence that summed it all up: a democracy must defend itself with force. Not with posts. Not with statements. With force. Michael Douglas is not just an actor. He is a proud Jew who isn’t ashamed, doesn’t apologize, and doesn’t bow his head. While all of Hollywood stays silent out of fear, he stands tall.

@nikillinit AI powered grill that supports Claude could be a thing

New post - why wearables are having a moment and how they might shape clinical care 1) the regulatory environment has meaningfully changed. ACCESS, TEMPO, and ADVOCATE. Provide new sandboxes for wearables to play in and get reimbursed 2) wearables start tracking from a healthy baseline, which introduces new ways to monitor disease progression 3) they introduce new data features like slope of change over time for normally static data points. We are starting to learn certain disease have different subtypes within them 4) wearables are likely the most defensible part of the consumer healthcare stack as the rest gets commoditized (lab testing, AI primary care, etc. Full post is here, @nikitabier said links aren’t being throttled anymore so I’m throwing it here: outofpocket.health/p/its-the-time…

What if… and hear me out now… instead of paying $5K per license for Claude with a dopey wrapper, you just pay $200 and use Claude?

Another year of training... what worked, what didn't, what had to change... A thread. Last year, I wrote the most detailed account I've ever published of how I train at 62. Well.. I'm turning 63 — now an update: what held, what broke, what surprised me, and the annoying realities of being an aging athlete.

Analysis of >50,000 public—AI chatbot health interactions @Copilot @Dominic1King @mooniean nature.com/articles/s4436…

Our paper on Subliminal Learning was just published in Nature! Last July we released our preprint. It showed that LLMs can transmit traits (e.g. liking owls) through data that is unrelated to that trait (numbers that appear meaningless). What’s new?🧵

@overton_news @ScottJenningsKY The only reason @abbydphillip has @davidhogg111 on her show is because he makes her feel smart.

This sliced through her ENTIRE narrative. Former Democrat Congresswoman Val Demings was dooming on CNN, practically claiming victory for the Iranian regime. Scott Jennings had heard enough and hit her with the one question Democrats hate answering. DEMINGS: “Negotiations failed. We don't know if they will start up again. There's a blockade. Yes, and it's effective. But it's not benefiting the United States...” “So where is the victory there?” JENNINGS: “Can I ask you a question?” “Do you want us to win? Do you want us to defeat these people?” DEMINGS: “I always want the U.S. to win...” JENNINGS: “Because it sounds to me like you are cheering for a bad outcome.” DEMINGS: “Could you define what win looks like?” JENNINGS: “Can you? Because all you can seem to do is define is negativity.” “Here’s what win looks like; these fanatics don’t get a nuclear weapon!” “PERIOD, FULL STOP.” “That’s a win.”

@agingroy @BMJ_Open @NBTiller BMJ is a 🤡 show.

ChatGPT 3.5 came out in November 2022. It's one of the models just tested in this @BMJ_Open study published today. @NBTiller's team found 50% of chatbot health answers were "problematic." Real finding, but for models that no longer exist. ChatGPT 3.5, Grok 2, Gemini 2.0, and Llama 3.3 are all now replaced. ChatGPT 3.5 is 5+ generations behind. In pharma, you'd never draw conclusions from a trial using a formulation replaced five times. But that's what happens when publication takes 1 or 2 years, and AI moves in weeks. We need a living benchmark for medical AI. @NBTiller, is a rolling version of this feasible?

@psychunseen The garbage part of this study is that it was performed on products that no longer exist because its over a year old. There are definitely problems today, but why not use models from TODAY. 🧐

Lots of mounting evidence that asking chatbots for medical advice is a bad idea. Within health topics where misinformation abounds, chatbots give problematic answers about half the time. Garbage in, garbage out. bmjopen.bmj.com/content/16/4/e…

@andrewarruda My bald head is very conducive to beanies, especially in the winter.

sabi in wired today! very proud. BCI is the future of all interfaces.

41 years ago today. The greatest round in boxing history. Hagler vs Hearns

Super excited to announce 🏥HealthAdminBench, a collaboration between @KineticSysAI and a great group of Stanford researchers from the labs of @drnigam @sanmikoyejo ! 🏥 HealthAdminBench is the first benchmark for LLM computer-use agents on solving administrative healthcare tasks which cost our economy $1 trillion/yr. 📊No model exceeds a 36% task success rate -- AI agents hate paperwork as much as humans! Full blog post here: kineticsystems.ai/blog/healthadm… (1/n)

Pretty crazy stat from IQVIA - for commercial patients, 70% of first time fills for branded drugs get rejected. This has been increasing over the years.