Scott McGrath

Scott McGrath

@smcgrath.phd

Biomedical Informatics PhD • CITRIS Health @UC Berkeley • FAMIA • Focusing on Informatics and AI in medicine • Linfield U. Grad • Missoula MT

10.8K followers462 following8.1K posts

Pinned

Scott McGrath
@smcgrath.phd

📣 New Paper Alert! Generalist physicians need AI literacy just as much as radiologists do. When I set out to build a new AI course, I couldn't find a medical curriculum guide that addressed this gap, so I decided to build one. Let me walk you through it. 🧵 1/4 #MedSky #MedEd

nature.comAI-PACE: a framework for integrating AI into medical education - npj Digital Medicinenpj Digital Medicine - AI-PACE: a framework for integrating AI into medical education
Scott McGrath
@smcgrath.phd

Huge improvement! Very happy with stripping out the extra ads.

Scott McGrath@smcgrath.phd

I'm very tempted to try out that vibe coding spywear removal from smart TVs. I put in some extra layers of protection on my PiHole after the LG case, but this seems like the better route.

Scott McGrath reposted

Ethan Linck
@elinck.bsky.social

last week to apply for full consideration—happy to chat informally if you are on the fence!

Ethan Linck@elinck.bsky.social

I'm hiring a postdoc to help build tools for inferring complex speciation histories with genomic data. strong computational skills are a must, but this is a training opportunity. 60K$+ salary, 2 years, details flexible. please reach out! (and repost) ad here: jobs.montana.edu/postings/53150

jobs.montana.eduPostdoctoral ResearcherThe Department of Ecology at Montana State University is seeking a Postdoctoral Research Scholar as part of NSF-funded research on speciation under alternating cycles of gene flow and isolation (fundi...
Scott McGrath
@smcgrath.phd

I'm very tempted to try out that vibe coding spywear removal from smart TVs. I put in some extra layers of protection on my PiHole after the LG case, but this seems like the better route.

Scott McGrath
@smcgrath.phd

USPS cut mail pickups in half across nearly 20k rural ZIP codes to save money, delaying postmarks by a day or more. Analysis found late-ballot rejection rates doubled or tripled in those counties during the 2026 primaries, hitting rural Trump-voting areas the hardest.

propublica.orgBallot Rejections Soared in Rural Counties After USPS Reduced Mail ServiceA ProPublica data analysis shows that delays resulting from the agency’s cost-cutting measures have endangered the validity of ballots mailed in rural areas. Counties that voted for Trump in 2024…
Scott McGrath
@smcgrath.phd

Prediction markets: $33 million in volume to tell you what the guys with Kalshi accounts believed last Tuesday. What a blight on society...

Josh Marshall@joshtpm.bsky.social

lol reminder that betting markets are extremely accurate measures of conventional wisdom

Scott McGrath reposted

Taniel
@taniel.bsky.social

Lula has been reelected president, Brazilian media has now called. An extraordinary comeback for the left-wing politician who trailed by 2 points in round 1 but managed to hold off Flávio Bolsonaro, the former president's son.

Scott McGrath
@smcgrath.phd

Stocked up for next weekend! I’m hoping 4 Costco bags and 1 box of full-sized candy bars gets us through the night. I think we went through 3.5 bags and 2/3rds of the full-sized box last year. We get a lot of kids in our neighborhood; it is a hotspot that people travel to. #Halloween

Scott McGrath reposted

Kate Starbird
@katestarbird.bsky.social

Go figure: Proposed bans of transgender athletes, like I-638 in WA, are framed as "protecting women" but that framing is misleading. This survey shows that people with more sexist views support bans at higher rates (79%) than people with less sexist views (28%). theconversation.com/are-transgen...

theconversation.comAre transgender sports bans really about protecting women? Survey data tells a different storyIn a recent national survey, Americans with the most negative views of women were the biggest proponents of transgender sports bans.

Scott McGrath reposted

Uncivil Servant
@noahreday.bsky.social

Looking at the abstract, they're looking solely at Part B expenses: diagnostic testing and referrals. Cost here doesn't include cost of new drugs. They also don't examine whether new drugs were necessary or even safe. There's no way to examine a difference in outcomes.

Scott McGrath@smcgrath.phd

🧪 A new study found care recommended by LLMs costs 3.2x more than plans from treating physicians ($314 vs. $97 per visit at Medicare rates). Across 200 real clinical encounters, AI ordered 5.6x more specialist referrals and nearly 2x as many new medications. #MedSky

ai.nejm.orgComparative Cost of Care Recommended by Large Language Models versus Treating PhysiciansLarge language models (LLMs) are increasingly consulted for clinical decision support, yet the cost of services or treatment they recommend has not been systematically evaluated. We evaluated 24 AI...

Scott McGrath reposted

Cat Hicks
@grimalkina.bsky.social

I've been kind of interested in this and whether it changes when you give models references about cost and testing decisions and it seems like it does when I experiment with this.

Scott McGrath@smcgrath.phd

🧪 A new study found care recommended by LLMs costs 3.2x more than plans from treating physicians ($314 vs. $97 per visit at Medicare rates). Across 200 real clinical encounters, AI ordered 5.6x more specialist referrals and nearly 2x as many new medications. #MedSky

ai.nejm.orgComparative Cost of Care Recommended by Large Language Models versus Treating PhysiciansLarge language models (LLMs) are increasingly consulted for clinical decision support, yet the cost of services or treatment they recommend has not been systematically evaluated. We evaluated 24 AI...
Scott McGrath
@smcgrath.phd

1 in 5 voters plans to use AI to research their 2026 midterm ballot, and 40% already get election details via AI search summaries, even though just 1% trust it as their most credible source. We've moved past deep fakes as the biggest election concerns.

nytimes.comA.I. and Its Dangers Are Embedded In the Midterm Elections (Gift Article)The midterms are the first elections in which artificial intelligence is integrated into every major social media platform. Efforts to blunt its harms are uneven.

Scott McGrath reposted

Noah Shachtman
@noahshachtman.bsky.social

The Russians hacked Jared's phone. Then Dutch cyberspies then reverse-engineered that hack, giving them unprecedented access to his world. Bananas story in @vanityfair.com

Vanity Fair@vanityfair.com

Jared Kushner’s Hacked Phone Revealed His Secret Negotiations—Including More Than $700 Million In Transfers

vanityfair.comThe Art of the Deal: Jared Kushner’s Hacked Phone Revealed More Than $700 Million in TransfersAn exclusive excerpt from ‘Princes and Thieves: Jared Kushner and Mohamed bin Zayed’s Secret Plot to Remake the Middle East’ examines how bank accounts linked the president’s son-in-law—and former sen...
Scott McGrath
@smcgrath.phd

🧪 A new study found care recommended by LLMs costs 3.2x more than plans from treating physicians ($314 vs. $97 per visit at Medicare rates). Across 200 real clinical encounters, AI ordered 5.6x more specialist referrals and nearly 2x as many new medications. #MedSky

ai.nejm.orgComparative Cost of Care Recommended by Large Language Models versus Treating PhysiciansLarge language models (LLMs) are increasingly consulted for clinical decision support, yet the cost of services or treatment they recommend has not been systematically evaluated. We evaluated 24 AI...
Scott McGrath
@smcgrath.phd

Physicians ordered zero diagnostic tests in 2 out of 3 visits, but every AI system ordered workups anyway. Models defaulted to reflexive screening panels like CBCs, metabolic panels, and HbA1c regardless of context. #MedSky

Scott McGrath reposted

Ethan Mollick
@emollick.bsky.social

By every measure, Opus 4.8 beats human experts as peer reviewers in a study of 124 Harvard social science papers covering a range from statistical analyses to formal models to qualitative work. As academics, we need to figure out how to adapt the practice of science to AI, and we need to do it fast

Scott McGrath
@smcgrath.phd

🧪 Once again, vaccines are amazing! I have no doubt anti-vax will end up in the dustbin of history, viewed as the modern equivalent of tobacco smoke enemas (once thought to revive drowning victims in the 1700s). Absolute lunacy in the face of actual miracles. #MedSky 🛟

Ana Marie Cox@anamariecox.com

Enormous news. A melanoma vaccine, one of the most deadly cancers out there. Science is good and helpful and there should be more of it. www.washingtonpost.com/health/2026/...

washingtonpost.comBreakthrough vaccine for the deadliest skin cancer cut risk by almost halfThe skin cancer vaccine is based on mRNA technology that helped end the coronavirus pandemic.
Scott McGrath
@smcgrath.phd

🧪 US healthcare generates ~8.5% of national greenhouse gas emissions, and 83% of the sector's footprint sits in the supply chain. A new npj Health Systems paper details how clinical informatics can track and cut waste in addition to fighting climate change! 🌎 #MedSky

nature.comHealth systems and climate change mitigation through an applied health informatics and sustainability perspective - npj Health Systemsnpj Health Systems - Health systems and climate change mitigation through an applied health informatics and sustainability perspective
Scott McGrath
@smcgrath.phd

🧪 Tracking ~30k brain scans across 12k kids shows the pace of cortical thinning can flag mental-health risk up to 4 years before symptoms appear. Delayed maturation at ages 9-11 predicted anxiety and depression at 13, though the signal is still too subtle for single-patient forecasts. 🧠

nature.comTeen depression and anxiety predicted by childhood brain developmentAnalysis of nearly 30,000 brain scans yields detailed timeline of how the brain matures — and hints at the consequences of disrupting it.
Scott McGrath
@smcgrath.phd

🧪 Global measles cases in kids under 5 had dropped nearly 80% from 1990-2019, then jumped 40% by 2023 to 15.6M cases (causing 118k deaths). New data shows 1 in 5 infants missed their first vaccine dose in 2023 (23.7M kids), pushing measles back up to the 6th leading cause of under-5 death. #MedSky 🛟

nature.comGlobal and regional burden of measles and vaccine coverage among children under 5 years, 1990−2023: a Global Burden of Disease Study - Nature MedicineUpdates to the Global Burden of Disease Study showed that although global measles incidence and mortality had declined in 1990−2019, its resurgence since 2020 reflects stalled vaccination progress…

Scott McGrath reposted

AMIA
@amia.org

Challenge Alert 💥 Registration is open for the National Library of Medicine’s SPARK Challenges—two prize competitions advancing AI-enabled discovery across PubMed/PMC & dbGaP. Total prizes up to $500,000. Learn more and register today: www.nlm.nih.gov/news/spark_c... #Informatics #MedSky

Scott McGrath
@smcgrath.phd

Congrats Bill, big year of achievements! Glad to see your impact being recognized! #MedSky

Bill Hersh@billhersh.bsky.social

A real honor to be inducted in the National Academy of Medicine this weekend! It is a year of awards for me and it would not have been possible without family, friends, colleagues, and students!

Scott McGrath
@smcgrath.phd

Over 950k US clinicians submitted 271M queries to OpenEvidence over 2.5 years, with 90% returning monthly. Usage skews heavily toward early-career doctors asking treatment questions. This is why I'm looking into the best way to help train med students in AI. Adoption is exceeding training. #MedED

jamanetwork.comEngaging With Medical Evidence in the Age of AIHealth care is, in part, an information science. Delivering optimal care requires clinicians to continually find, evaluate, and apply an expanding body of medical knowledge when caring for patients.…
Scott McGrath
@smcgrath.phd

Interesting: Epilepsy devices track brain signals, but they usually fail to capture what the patient felt. A small study (n=13) out in Nature Medicine, used a GPT-4o-mini agent to chat with patients after seizure alerts, turning their replies into training labels for fine-tuning. 🧪 🧠

nature.comConversational brain AI for adaptive epilepsy care - Nature MedicineA small study shows that patients can help fine-tune their own seizure detectors by chatting with an AI agent. However, the more difficult challenge ahead is to build AI that responds appropriately.
Scott McGrath
@smcgrath.phd

Anthropic is giving six METR researchers employee-level access to audit its frontier models and risk practices. The setup works like resident bank examiners: Anthropic can't veto reports, and METR can publicly challenge any redactions.

techcrunch.comAnthropic is letting METR researchers look under the hood with employee-level access | TechCrunchMETR will publish its own assessments of Anthropic’s risks, incidents, and practices, and the frontier AI model maker has given up the right to block them.
Scott McGrath
@smcgrath.phd

OpenAI has been shaking up the math world by dropping 350+ math findings (including a a Navier-Stokes proof). Fields Medalist Dr. Terry Tao was an early AI adopter, but says this push to benchmark models is strip-mining years of PhD work and leaving dense proofs with no human author to question.

nytimes.comTerry Tao Is on the Front Lines of the Debate Over AI and the Future of Math (Gift Article)The Fields Medalist and celebrated mathematician is on the front lines of a debate about artificial intelligence and the future of his field.
Scott McGrath
@smcgrath.phd

I’m frustrated with people trying to turn Shannon OBrien’s resignation into a scandal when the bigger story should be that Blankhead isn’t a very good candidate. Particularly when the “evidence” stated against Shannon comes from non-neutral parties.

Scott McGrath
@smcgrath.phd

Ballot officially dropped off! I already know I will be disappointed in the Senate race, but things are looking up for Sam and I-195 (the anti-dark money initiative). Make sure to have a plan in place, don’t rust the polls, get out there and VOTE!

Scott McGrath reposted

public health guy
@publichealthguy1.bsky.social

@kkjetelina.bsky.social nails it here when she notes misinfo is not just tech/proliferation problem—it's fundamentally a symptom of too many people being unable to get the info they need from sources they trust. that's why (imo) we've gotta come at the problem from multiple angles.

STAT@statnews.com

STAT's new trust-in-science reporter, Nick Florko, asks Katelyn Jetelina about how she decides whether to address misinformation: #STATSummit

Scott McGrath
@smcgrath.phd

Tagging for 🧪 #AcademicSky

Katie Spoon@kspoon.bsky.social

As a PhD student designing my first survey, I added what I thought was a throwaway question at the end. To my surprise, thousands of academics answered it! The paper is out today: doi.org/10.1093/pnas...