Lightnews — Scholar-powered news

Scott McGrath

@smcgrath.phd

9.7K followers 400 following 5.4K posts

Biomedical Informatics PhD • CITRIS Health @UC Berkeley • FAMIA • Focusing on Informatics and AI in medicine • Linfield U. Grad • Missoula MT https://citris-uc.org/people/person/scott-mcgrath/

citris-uc.org

Posts Media Videos Starter Packs

Pinned

Scott McGrath @smcgrath.phd · 16d

Decided to try something new out. If you like the ML Feed: Trending, try out the new ML Feed: Blend!

Instead of just showing the trending ML posts from the past 24 hours, this will show new posts mixed in with popular posts. It should provide a little more variety!

1 3

Reposted by Scott McGrath

Yuri Quintana @yuriquintana.com · 1d

New paper on AI agency.
LIMI: Less is More for Agency shows that building autonomous AI agents doesn’t require massive datasets, just carefully curated demonstrations.
With only 78 samples, LIMI outperforms models trained on 10,000+.
arxiv.org/abs/2509.17567

#AI #Agency #AutonomousAgents #LIMI

LIMI: Less is More for Agency

We define Agency as the emergent capacity of AI systems to function as autonomous agents actively discovering problems, formulating hypotheses, and executing solutions through self-directed engagement...

arxiv.org

1 1

Scott McGrath @smcgrath.phd · 22h

Wolters Kluwer has launched UpToDate Expert AI, a GenAI version of its clinical decision support tool.

The system is designed to prevent AI hallucinations by exclusively drawing from UpToDate's peer-reviewed content, not the open web.
#MedSky #MedAI

Wolters Kluwer jumps into the AI market, rolls out gen AI version of UpToDate

Wolters Kluwer Health launched a version of UpToDate supercharged with artificial intelligence as competition in the generative AI healthcare market heats up. | Wolters Kluwer launched a version of Up...

www.fiercehealthcare.com

Scott McGrath @smcgrath.phd · 22h

Generative AI is evolving beyond simple tasks. New reasoning and agentic models can handle complex, multistage problems in medicine. There is the potential to improve healthcare, but careful validation and thoughtful adoption are critical.

#MedSky #MLSky #MedAI

Generative artificial intelligence in medicine - Nature Medicine

This Review summarizes recent technical advancements in generative AI, outlines how new models might improve healthcare and discusses validation approaches—using lessons from recent successes and…

www.nature.com

Scott McGrath @smcgrath.phd · 22h

Digital twin models show promise for predicting neutropenia in AML patients on venetoclax and azacitidine. While patient-specific accuracy varies, continuous updating over time significantly improves the model's predictive power.
#MedSky #MedAI

Digital twin models for predicting venetoclax and azacitidine-induced neutropenia in patients with acute myeloid leukemia - npj Digital Medicine

npj Digital Medicine - Digital twin models for predicting venetoclax and azacitidine-induced neutropenia in patients with acute myeloid leukemia

www.nature.com

Reposted by Scott McGrath

Simon Willison @simonwillison.net · 23h

Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced

What about when engineers at the top of their game use AI tools responsibly to accelerate their work?

I propose "vibe engineering"!

simonwillison.net/2025/Oct/7/v...

Vibe engineering

I feel like vibe coding is pretty well established now as covering the fast, loose and irresponsible way of building software with AI—entirely prompt-driven, and with no attention paid to …

simonwillison.net

27 30 210

Scott McGrath @smcgrath.phd · 23h

They are great!

There is a bit of tension due to some of the class 2 (ones w/ throttles) bikes playing a a little too loose with the speed limits.

However, I'm thrilled with my class 3 cargo bike.

Scott McGrath @smcgrath.phd · 23h

Sam Altman says ChatGPT has hit 800M weekly active users

techcrunch.com/2025/10/06/s...

Sam Altman says ChatGPT has hit 800M weekly active users | TechCrunch

OpenAI CEO Sam Altman said Monday at Dev Day that ChatGPT has reached 800 million weekly active users, marking an increase of adoption among consumers, developers, enterprises, and governments.

techcrunch.com

1 1

Scott McGrath @smcgrath.phd · 23h

Quite a blunt assessment for what Bari Weiss signed up for.

The Linda Yaccarino without the advertising connections analogy seems fairly on the mark.

Memo to Bari Weiss Re: CBS News: You’re doomed

Bari Weiss has sold her company, The Free Press, to manage the decline of broadcast news at CBS. How many ways can it go wrong?

www.theverge.com

8 41

Reposted by Scott McGrath

Caitlin Gilbert @caitlingilbert.bsky.social · 1d

🤳EXCLUSIVE: We analyzed real data from over 800 U.S. TikTok users to reveal how the app transforms casual users into power users.

Extremely proud of this one, over a year in the making w/ @richardsima.bsky.social Leslie Shapiro @asteckelberg.bsky.social @ence.bsky.social

wapo.st/4obueib

Here’s how TikTok ‘power users’ end up scrolling 4 hours a day or more

Washington Post reporters analyzed data from over 800 TikTok users to learn how the app turns people into power users, some of whom spend hours per day scrolling.

wapo.st

10 60 140

Scott McGrath @smcgrath.phd · 1d

Chilly Fall morning

Setting Moon over the mountains with fog clinging to a field

Scott McGrath @smcgrath.phd · 1d

tagging for #MedSky #MedEd #MedAI

Jonathan H Chen MD PhD @jonc101.bsky.social · 1d

@stanforddeptmed.bsky.social Biomedical Informatics Research Colloquia
“Applied Intelligence: Integrating AI Technologies Into Medical Education”
Laurah Turner, PhD.

Thursday, October 9th, 2025
12:00 to 1:00 pm PST

stanford.zoom.us/j/9788759601...

Webinar ID: 978 8759 6012
Passcode: 420642

Reposted by Scott McGrath

Taylor Nichols, MD FACEP @tnicholsmd.com · 1d

What is a song that likely doesn't make anyone's top lists, but that you think is a perfect song, or absolutely beautiful, or that you could listen to on repeat?

52 2 27

Scott McGrath @smcgrath.phd · 1d

Pearl Jam - The Fixer

Smashing Pumpkins - Mayonaise

1 5

Scott McGrath @smcgrath.phd · 1d

Pinning any of these feeds will give you tabs on your home page you can scroll through and give you more to explore and engage with.

Scott McGrath @smcgrath.phd · 1d

You can even build your own feeds, which could be as simple as tracking posts from a single account on a single topic, or more advanced topics like this Machine Learning feed I built (I have a medical AI one in the works)

Scott McGrath @smcgrath.phd · 1d

Popular with Friends - posts that your friends have liked

Scott McGrath @smcgrath.phd · 1d

MedSky - the tent pole medical feed with a ton of medical specialty feeds ( open up @medsky.social and scroll to their feeds tab to see all of the options)

1 2

Scott McGrath @smcgrath.phd · 1d

Quite posters - this feed highlights people you follow that don’t post frequently so you don’t miss them

1 1 3

Scott McGrath @smcgrath.phd · 1d

For You - this is better than the default Discover feed, it surfaces things you might enjoy based off the post you like:

Scott McGrath @smcgrath.phd · 1d

Twitter/X is more like an automatic car: the algorithm there does a lot of the heavy lifting, but you have a lot less control.

Bluesky is more of manual car: you have to set more thing up, but then have much more control over things. The key is pinning some good feeds.

Here are my suggestions:

2 7

Scott McGrath @smcgrath.phd · 1d

LLMs are often ineffective at processing time series data, a significant limitation for medical applications. This preprint introduces OpenTSLM, enabling LLMs to reason over medical time series through a native modality.
#MedSky #MedAI #MLSky

OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data

LLMs have emerged as powerful tools for interpreting multimodal data. In medicine, they hold particular promise for synthesizing large volumes of clinical information into actionable insights and…

arxiv.org

2 5

Scott McGrath @smcgrath.phd · 1d

🧪 Amid concerns about AI and funding, the value of STEM education remains.

Science is the process of building and refining models, not just knowing facts. AI is a tool in this process, handling data while humans ask the fundamental questions.

How to Get Your Kids Into STEM Even When Its Future Is Uncertain

Thinking about science and technology in terms of return on investment misses the point. Here’s what kids really need to know.

www.wired.com

1 5

Scott McGrath @smcgrath.phd · 1d

🧪 AI agents are emerging as tools to automate multi-step research tasks, from data curation to hypothesis generation. While they can act as research assistants, they are not yet autonomous and require significant human oversight to mitigate risks.
#MLSky

How AI agents will change research: a scientist’s guide

Researchers are increasingly turning to artificial-intelligence tools that can handle complex, multi-step processes.

www.nature.com

Scott McGrath @smcgrath.phd · 1d

AI skills are improving unevenly due to a "reinforcement gap." Tasks like coding, which can be automatically tested at scale, improve rapidly via reinforcement learning. Subjective tasks like writing, lacking clear metrics, progress more slowly.
#MLSky

The Reinforcement Gap — or why some AI skills improve faster than others | TechCrunch

AI tasks that work well with reinforcement learning are getting better fast — and threatening to leave the rest of the industry behind.

techcrunch.com

Scott McGrath @smcgrath.phd · 1d

🧪 #MedSky 🧬💻 Direct link to the study: www.nature.com/articles/s41...

Polygenic and developmental profiles of autism differ by age at diagnosis - Nature

A study of several longitudinal birth cohorts and cross-sectional cohorts finds only moderate overlap in genetic variants between autism that is diagnosed earlier and that diagnosed later, so they…