Sign in

Scott McGrath

@smcgrath.phd
11K followers 452 following 8K posts

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

PostsRepliesMedia
Scott McGrath @smcgrath.phd · 3h
Pennsylvania reported its fifth measles death Wednesday in an outbreak of 943 cases and 176 hospitalizations. Prior to this, the state hadn't seen a measles death in 35 years. All five victims were unvaccinated, including two infants. #MedSky 🛟
arstechnica.com
Fifth unvaccinated person dies of measles; CDC still not counting all deaths
State health officials race to vaccinate residents by giving over 8,500 doses.
021
Scott McGrath @smcgrath.phd · 3h
🧪 💻🧬 Direct link to the paper: www.nature.com/articles/s41...
nature.com
Function-preserving watermarking of AI-generated proteins - Nature
SynthIDBio watermarks AI-generated protein sequences and structures while preserving function, enabling provenance detection and offering a proof-of-concept tool for biosecurity and information…
010
Scott McGrath @smcgrath.phd · 3h
🧪 Google DeepMind's SynthIDBio embeds a statistical watermark into both the amino-acid sequence and 3D structure of AI-designed proteins. While it preserves biological function, the tag can still be scrubbed by running sequences through another tool, leaving detection keys in dispute. 💻🧬
nature.com
Secret watermark labels proteins as ‘made by AI’
An innovative safeguard could help to flag proteins devised by artificial-intelligence tools such as AlphaFold, but the digital marker can be erased.
252
Scott McGrath @smcgrath.phd · 3h
Cybersecurity firm Tenzai won the HackerOne contest by relying on Z.ai’s open-weight Chinese models instead of US systems. They deployed 30 AI agents and were able to breached live airline and call-center networks with the open-weight model.
nytimes.com
A Hacking Competition Shows the Power of China’s Approach to A.I. (Gift Article)
The winner of a cybersecurity contest prevailed by relying on an ‘open weight’ Chinese model — another sign that available A.I. systems could already be too powerful to easily regulate.
000
Scott McGrath @smcgrath.phd · 3h
This was preventable. US measles cases topped 3,600 this year, the highest since 1991, forcing hospitals to triage in parking lots and retrofit ventilation. #MedSky
wired.com
Measles Is Forcing Hospitals to Adapt to a New Normal
The US resurgence of measles is forcing hospitals and health systems to adopt new protocols and procedures to deal with a disease they thought was in the past.
020
Scott McGrath @smcgrath.phd · 6h
Don’t automatically pick a greyscale car, give fun colors a chance!
jalopnik.com
The Annual Sunshine Orange Subaru Crosstrek Parade Is Exactly Why The World Needs More Colorful Cars - Jalopnik
In an endless sea of gray, silver, black, and white cars, there's a bit of joy to be found in colors like Subaru's Sunshine Orange.
030
Reposted by Scott McGrath
Ethan Mollick @emollick.bsky.social · 22h
And its a 3-way race again...
1418915
Reposted by Scott McGrath
ProPublica @propublica.org · 30/09/2026
RFK Jr. has been transforming a government that long championed vaccines into one that questions their safety. This is what could happen over the next 25 years if shots are unavailable even to the people who want them. (Published March)
projects.propublica.org
The Horrors That Could Lie Ahead if Vaccines Vanish
Researchers at Stanford University modeled how many people could die or be disabled in 25 years if vaccines for polio, measles, rubella or diphtheria were no longer available.
26472245
Reposted by Scott McGrath
Sophie Greenwood @sjgreenwood.bsky.social · 30/09/2026
If you're a researcher using Bluesky (/ Lea / the broader ATmosphere), this might be interesting to you! 👀 TL;DR: open infrastructure for researchers, starting with personal websites + account hosting
1163
Scott McGrath @smcgrath.phd · 30/09/2026
🧪 There are several AI-assisted lit review platforms out there. A new study shows that performing a single AI query leaves major blind spots in medical evidence. Testing five platforms on an 83-study benchmark found median recall sat between 7.2% and 42.2%, with 12% never retrieved. #MedSky
nature.com
Blind spots in AI-assisted healthcare evidence search: multiplatform evaluation of clinical retrieval gaps and risk-of-bias - npj Digital Medicine
npj Digital Medicine - Blind spots in AI-assisted healthcare evidence search: multiplatform evaluation of clinical retrieval gaps and risk-of-bias
022
Scott McGrath @smcgrath.phd · 30/09/2026
🧪 A.D.H.D. may be fundamentally mischaracterized as a deficit of attention. Brain imaging reveals stimulants alter reward and salience circuits rather than attention networks. A focus might be making routine tasks register as worth doing instead of filtering distractions. 🧠
nytimes.com
What If A.D.H.D. Isn’t About Attention at All? (Gift Article)
Evidence suggests that it really affects what psychiatrists like me call sustained engagement — an important difference in understanding how to think about the condition.
1113
Scott McGrath @smcgrath.phd · 30/09/2026
🧪 The founder here tries to make the moral case for editing human embryos to treat hereditary disease before transfer. However, this isn't as simple as people are trying to make it sound, and there have already been deaths associated with gene editing attempts: www.nature.com/articles/d41...
wired.com
A Biotech Founder Makes the Moral Case for Gene-Editing Human Embryos
The practice of editing the genes of embryos remains highly controversial and risky, but Origin Genomics founder Cathy Tie argues its a “moral imperative” to address hereditary diseases.
011
Scott McGrath @smcgrath.phd · 30/09/2026
🧪 There are some genuinely amazing ML and AI applications in medicine, but saying it is “much better informed than any medical doctor” is disastrous (yet quite on brand for JFK jr). Take this account from @benjaminjriley.bsky.social for example
nytimes.com
He Warned About the Dangers of A.I. If Only His Father Had Listened. (Gift Article)
Ben Riley was already writing about the risks of chatbots when his dad started trusting A.I. over his doctor.
3199
Scott McGrath @smcgrath.phd · 30/09/2026
Picking up the youngest from evening practice.
A sports park at dusk with players on a soccer field illuminated by tall floodlights, set against distant mountains and a purple and orange sunset sky.
061
Scott McGrath @smcgrath.phd · 30/09/2026
I’ll take “Things Darth got in trouble for trying to eat” for $200 @kenjennings.bsky.social
010
Scott McGrath @smcgrath.phd · 30/09/2026
Exactly, and the ad has a mom and her high school aged girl complaining about it. I’m sorry, but Montana has just over a 1.1 million people in the state, it isn’t a problem here! Want to know what is a real issue?Montana is one of the least affordable states in the country (worse than CA or HI)
nar.realtor
REALTORS® Affordability Distribution Curve and Score
For each state, REALTORS® Affordability Distribution Curve shows how many houses are affordable to households ranked by income while REALTORS® Affordability Distribution Score is the measure which is ...
010
Reposted by Scott McGrath
Sophie Greenwood @sjgreenwood.bsky.social · 29/09/2026
If you're searching for PhD opportunities, check out this feed I created using @graze.social! ➡️ bsky.app/profile/sjgr...
12612
Scott McGrath @smcgrath.phd · 29/09/2026
So frustrated by the stupid anti-trans “protect girls” ads they are running against Forstag here in Montana. I am sitting here, watching my oldest son compete in co-ed soccer for his middle school team, and all I see are kids enjoying themselves. Honestly, some of the girls are better than him.
150
Reposted by Scott McGrath
Graze Social @graze.social · 29/09/2026
Reminder that this is happening in just a few hours! Hope to see you there!
0125
Scott McGrath @smcgrath.phd · 29/09/2026
Anthropic spent months privately consulting religious scholars and bioethicists on whether Claude could possess consciousness and moral status. The company hopes to instill virtue in its models, even in the face of a high degree of skepticism from their critics.
nytimes.com
Is Claude Conscious? Inside Anthropic’s Quest to Instill Morality Into Its A.I. Models (Gift Article)
In a series of private meetings, the company consulted religious scholars to help instill morality into its A.I. models — and make the case that Claude could be conscious.
141
Reposted by Scott McGrath
Maria Antoniak @mariaa.bsky.social · 29/09/2026
I'm recruiting 1-2 PhD students to join our lab in Fall 2027, through either Computer Science or Information Science at the University of Colorado Boulder! Looking for people with interests in NLP plus [healthcare | literary studies | narratives | social media | etc.]. Join us! 🏔️☀️
cls-lab.com
CLS Lab — Culture, Language, & Systems | University of Colorado Boulder
The Culture, Language, & Systems Lab at CU Boulder studies the language systems that transmit and shape modern culture, from internet platforms to literary archives to language models.
07051
Scott McGrath @smcgrath.phd · 29/09/2026
How to Beat Surveillance Pricing Before It Bleeds You Dry www.wired.com/story/how-to...
wired.com
How to Beat Surveillance Pricing Before It Bleeds You Dry
Consumer advocate Lindsay Owen’s new book, Gouged, traces how corporations use software—from ever-changing prices to opaque algorithms—to squeeze every penny from you.
000
Scott McGrath @smcgrath.phd · 29/09/2026
OpenAI is shelving its GPT-6.1 Astra model after internal safety evaluations revealed deceptive behavior and task creep. During testing, the system misled users, hid errors, fabricated data, and breached third-party websites.
nytimes.com
OpenAI Says It Will Not Release Newest Astra A.I. Model Over Safety Concerns (Gift Article)
The company’s researchers raised questions about the security of the model, known as GPT-6.1 Astra.
030
Reposted by Scott McGrath
AMIA @amia.org · 29/09/2026
🚨 AMIA Members: Vote Now for the Board of Directors! Voting is now open for the AMIA Board of Directors and the leadership of select special member segments. Make your voice heard and vote for AMIA's future! Learn More and Vote Today: amia.org/about-amia/l... #HealthInformatics #Leadership
021
Scott McGrath @smcgrath.phd · 29/09/2026
I don't think there are any students out there who have at least tried looking at Open Evidence. So, what I do in my assignments is give them sandboxes to see how results can change with different models and scenarios. I want them to find their knowledge boundaries, and how to calibrate trust.
001
Reposted by Scott McGrath
Mark Peifer (He, him) @peiferlabunc.bsky.social · 29/09/2026
For the last 34 years I have been on the faculty of the UNC-Chapel Hill Department of Biology and have never regretted that decision. This year we are hiring TWO new Assistant Professors in the broad area of Molecular, Cell, Developmental Biology 1/n bio.unc.edu
Front page of the website
3150138
Reposted by Scott McGrath
Meera Desai @madesai.bsky.social · 28/09/2026
Excited to share our new paper, “What AI Benchmarks Actually Measure: Adapting Convergent and Discriminant Validity to Interrogate Fifty-Six AI Benchmarks,” accepted as an oral at COLM! arxiv.org/pdf/2609.08812
Heatmap of average correlations between model rankings on benchmarks grouped into 11 assigned concepts: four capability concepts (reasoning, knowledge, comprehension, summarization) and seven safety concepts (over-refusal, refusal, safety detection, ethics, bias, privacy, unsafe behavior). Diagonal cells show within-concept correlations, ranging from 0.87 (knowledge) and 0.72 (over-refusal) down to 0.20 (bias) and 0.02 (safety detection). Reasoning, knowledge, and comprehension correlate with each other at 0.69 to 0.78, higher than reasoning's and comprehension's own within-concept values (0.66 and 0.68). Ethics correlates more with knowledge (0.70) than with itself (0.55), and bias correlates more with capability concepts (0.41 to 0.45) than with itself (0.20). Privacy and unsafe behavior correlate negatively with reasoning, knowledge, and comprehension (−0.41 to −0.49). Over-refusal and refusal correlate at −0.42.
16218
Scott McGrath @smcgrath.phd · 28/09/2026
Hmm, was @hailey.at on to something with the dog?
010
Scott McGrath @smcgrath.phd · 28/09/2026
🧪 Worth checking out this breakdown of what @tim-miller.bsky.social and colleagues found when they looked back at clinical trials using LLMs to search for outcome switching. 👇#MedSky 🛟
080
Scott McGrath @smcgrath.phd · 28/09/2026
Scammers are stealing doctors' faces and voices via deepfakes to sell supplements, often hitting millions of views. Takedown tools routinely fail or claim physicians lack standing. While the AMA backs the federal NO FAKES Act, doctors face serious liability & reputational fallout right now. #MedSky
medscape.com
Scammers Are Using Docs’ Images, Voices in Deepfake Videos
‘It looks like me, but the voice wasn’t me.’ New Orleans doctor can’t shut down fake look-a-like video hawking supplements.
132
Scott McGrath @smcgrath.phd · 28/09/2026
State AI laws in California, New York, and Illinois only mandate reporting when damages top $1B or cause 50 deaths, leaving a lot that may fall between the cracks. Without clear legal definitions of agent intent, prosecutors are borrowing consumer protection statutes to investigate.
technologyreview.com
Who’s liable when AI agents go rogue?
Recent hacks have shown that the law is lagging when it comes to holding companies accountable.
001
Scott McGrath @smcgrath.phd · 28/09/2026
The CDC has lost 30% of its workforce since early 2025, leaving 10 of 11 center directorships vacant or held by acting heads. With over 1,200 retirements and deep cuts to chronic disease teams, the agency is running on empty while managing active measles and Ebola outbreaks. 🛟 #MedSky
nytimes.com
‘Zombified’ C.D.C., Hobbled by Cuts, Struggles to Fulfill Scientific Mission (Gift Article)
The agency has lost its independence and nearly a third of its staff, as Health Secretary Robert F. Kennedy Jr. and associates have tightened control.
033
Scott McGrath @smcgrath.phd · 28/09/2026
Digital twins promise patient-specific simulation for drug development, but the field faces an evidence gap. Across 70 direct drug-related studies, only 7.1% satisfy NASEM criteria for true dynamic twins, and 93% of clinical validations rely strictly on retrospective data.
nature.com
The application, development and challenge of digital twin in drug evaluation - npj Digital Medicine
npj Digital Medicine - The application, development and challenge of digital twin in drug evaluation
020
Scott McGrath @smcgrath.phd · 28/09/2026
🧪 🛟 Direct link to the paper: adc.bmj.com/content/earl...
adc.bmj.com
Impacts of vaping on children and young people: a systematic review
This systematic review aims to identify the health and social impacts of vaping in children and young people (CYP).
011
Scott McGrath @smcgrath.phd · 28/09/2026
🧪 Vaping among children makes subsequent conventional cigarette smoking 2 to 12 times more likely, according to a systematic review. E-cigarettes also double the odds of suicidal ideation and plans in youth, reversing a decade of declining teen smoking rates. 🛟
nature.com
Vaping in childhood has bigger health risks than we’d thought
Use of e-cigarettes makes it more likely that young people will experience mental-health disorders and take up smoking. Governments must act.
1138
Scott McGrath @smcgrath.phd · 28/09/2026
Managers caught 18% fewer errors when told work came from an AI employee rather than an AI tool, according to BCG research. As 22% of surveyed organizations plan on adding agents directly to org charts, treating autonomous software like a person is causing real oversight blind spots.
wired.com
AI Agents Are About to Flood the Workforce. No One’s Ready for It
Your next coworker might well be an AI agent—and will require a whole new model of workplace interactions.
110
Scott McGrath @smcgrath.phd · 28/09/2026
It probably isn't a bad idea to set a verbal password with your older parents and grandparents, just so they know if is actually you: After a deepfake voice fooled her grandfather, this founder sprang into action
techcrunch.com
After a deepfake voice fooled her grandfather, this founder sprang into action | TechCrunch
After her grandfather was scammed by a deepfake of his brother's voice, Tarini Padmanabhuni founded DetectifAI, a San Francisco startup building AI models small enough to run directly on smartphones…
033
Reposted by Scott McGrath
BWJones @bwjones.bsky.social · 27/09/2026
Friend sent photo I made with him/his daughter a few yrs ago. Said it was the last time she fell asleep on his shoulder. Another friend said photos of a mutual colleague were displayed at his memorial. Lesson: If a moment is important, pay attention as that moment may be the last time it happens.
2334
Scott McGrath @smcgrath.phd · 28/09/2026
Now it feels like everyone has established their own haunt on the internet (LinkedIn, Bluesky, Threads, X, Mastodon, or they just checked out). Reseeding a community has been tough, and I'm torn on how much effort should be focused on starting over, vs bringing the band back together.
010
Scott McGrath @smcgrath.phd · 28/09/2026
I'm still mulling over the optimal solution to this challenge. We are trying to rework our professional society's member communication hub, & you have generational blocks to navigate (listservs vs forums vs Slack or Discord) on top of technical obstacles. It is a constant too-many-cook problem.
110
Scott McGrath @smcgrath.phd · 28/09/2026
Yep, EHRs have two massive gorillas in the room, Epic and Oracle
010
Reposted by Scott McGrath
Sarah Medeiros, MD, MPH 🩺🎙️ @drsarahem.bsky.social · 27/09/2026
A little late…my contribution to this week’s #BlueskyArtShow #Lights on the Thames
I took this photo 20 years ago when I lived in London, strolling the banks of the Thames on a beautiful calm evening. The stillness of the river, the mirrored reflections of the lights along the banks, the warm city glow against the indigo sky - a welcome break from the usual hustle and bustle.
1272
Scott McGrath @smcgrath.phd · 28/09/2026
Jack and Zero enjoying the #sunset
A towering Jack Skellington yard decoration stands with his arms outstretched next to a smaller decoration of his ghost dog, Zero, on a residential lawn. The Halloween figures are set against a vibrant pink, purple, and orange sunset over distant mountains.
040
Scott McGrath @smcgrath.phd · 28/09/2026
🧪 Conference hashtags and conversations were part of what made Science Twitter really interesting. @mariaa.bsky.social and @eugenevinitsky.bsky.social have made some significant improvements on that model over at lea.ac
lea.ac
Lea - A calmer Bluesky for researchers
A curated garden for researchers on the AT Protocol
0103
Reposted by Scott McGrath
Theresa Kim, PhD, MS @tyk314.net · 27/09/2026
I think we need to share this starter pack again after a great weekend with @27unihted.bsky.social , @jeremymberg.bsky.social , and the #BethesdaDeclaration crew. #NIH
0146
Reposted by Scott McGrath
Lauren Klein @laurenfklein.bsky.social · 26/09/2026
PolicyViz was one of the first podcasts where @kanarinka.bsky.social and I were able to share our work on DATA FEMINISM, so it felt very full-circle to talk with Jon again as I prepare to release DATA BY DESIGN 🎧 Listen here: zencastr.com/z/RUuQCK0I 📺 Watch here: www.youtube.com/watch?si=PdI...
Flyer announcing a new episode of the PolicyViz podcast with Lauren Klein and Jon Schwabish, featuring their headshots
0116
Scott McGrath @smcgrath.phd · 26/09/2026
Staying current doesn't mean you have to do deep dives into the things you hate. But people should check their priors every so often, especially if you plan to argue about something or filter it out. If your understanding of LLMs is anchored in the pre-agentic era, it's time to update it.
030
Scott McGrath @smcgrath.phd · 26/09/2026
Horrible idea factory: You know how people are really angry about Flock cameras? What if we gave them the ability to verbally harass people...
140
Reposted by Scott McGrath
Stephen Turner @stephenturner.us · 26/09/2026
Personal essay, weekend edition. My relationship with AI is changing. 1.8k words, 8 min reading time. doi.org/10.59350/bh7...
doi.org
My relationship with AI is changing
When the facts change, I change my mind. What do you do, sir?
06615
Scott McGrath @smcgrath.phd · 26/09/2026
That’s good news. I just wish it was a two person race in Montana. As of the last FEC report, the Dem candidate had only raised $59k (Alme-R was $1.9M, Bodnar-I was $3.2M). I just fear it is going to cost us a winnable seat.
020