Scott McGrath

Tagging for #MedSky

A Word to the Wise: Don’t Trust A.I. to File Your Taxes

This shouldn't be surprising, but do resist the temptation: LLMs are not ready to file your taxes for you.

Factors for Patient Trust and Acceptance of Medical Artificial Intelligence This survey study examines the associations of patient trust in and choice of medical scenarios involving artificial intelligence with receiving information on governance mechanisms, clinician…

What actually makes patients trust an AI diagnosis?
A survey of 3k adults asked people to choose between hypothetical AI diagnosis scenarios and pick a "visit" from the options presented. AI performance accuracy boosted visit preference by 32.5%, while a clinician in the loop added 18.4%.
#MedSky

RFK Jr.’s anti-vaccine policies are "unreviewable," DOJ lawyer tells judge The startling claim came amid a lawsuit from the American Academy of Pediatrics.

A DOJ lawyer just told a federal judge that Health Secretary RFK Jr.’s vaccine policies are "unreviewable". This claim of limitless authority comes as the AAP and medical groups sue to block unilateral changes to the CDC's childhood vaccine schedule.
#MedSky 🛟

ChatGPT Health performance in a structured test of triage recommendations - Nature Medicine A stress test of ChatGPT Health triage revealed missed high-risk emergencies and inconsistent activation of suicide-crisis safeguards, raising safety concerns for consumer-scale deployment.

Direct link to the Nature study looking at ChatGPT Health: www.nature.com/articles/s41...

The era of Doctor AI is already here Artificial intelligence is opening access to medical information in an entirely new way.

40M people ask ChatGPT health questions daily.
While OpenAI promotes that GPT-5 hits 99% triage accuracy, a Nature study found older models under-triaged 50% of emergencies. Convenience is winning, but pushing forward before these systems are validated is irresponsible.
#MedSky

Pricing models for diagnostic AI based on qualitative insights from healthcare decision makers - npj Digital Medicine npj Digital Medicine - Pricing models for diagnostic AI based on qualitative insights from healthcare decision makers

Pricing for diagnostic AI is often a black box of technical metrics that don't fit hospital budgets. A new study (n=17) found a 76% resistance to pure usage models. Turns out, leaders want "clinically legible" units, like per patient or per case, to make AI both predictable and scalable.
#MedSky

NIH reneges on recognizing union for early career researchers The biomedical research agency says trainees in its labs are not “employees”

Nearly 5,000 early-career researchers at the NIH just saw their union recognition pulled. The agency now argues these fellows are trainees, not employees, despite a ratified contract from 2024. It’s a massive shift that puts raises and health insurance for 5,000 scientists in limbo.

AWS launches a new AI agent platform specifically for healthcare | TechCrunch AWS is launching Amazon Connect Health, an AI agent platform that will help with patient scheduling, documentation, and patient verification.

AWS launches a new AI agent platform specifically for healthcare

techcrunch.com/2026/03/05/a...
#MedSky

991.1 C2

Cluely CEO Roy Lee admits to publicly lying about revenue numbers last year | TechCrunch The $7 million in annual recurring revenue that Cluely CEO Roy Lee shared last summer was a lie, its founder and CEO Roy Lee admitted on Thursday on X.

Is there no honesty in the cheating racket anymore?

TriZetto confirms 3.4M people's health and personal data was stolen during breach | TechCrunch Health tech giant TriZetto has confirmed that more than 3.4 million people had personal and health information stolen in a 2024 cyberattack, which the company failed to detect for almost a year.

TriZetto confirms 3.4M people’s health and personal data was stolen during breach

techcrunch.com/2026/03/06/t...

An all time classic. Love Guards!

Waiting for the roads to get swept for gravel to get mine back on the road.

Absolutely lovely!

Guards Red?

Enjoy the ride!

It tends to be a mix of all three.

There is a whitelist of a little over 1600 people with a RegEx for AI/ML terms that picks up on their posts, matches them, and then sorts them depending on the feed (blend: new & trending, trending only, or personalized).

📌 New feed alert!

Excited to announce - ML Feed: For You

It is the AI/ML same feed, but now powered by @graze.social's personalization algorithm.

Please check it out, and let me know what you think!

Another new feature will be rotating posts.

These posts will show up with a repeat emoji 🔁

Thanks for checking it out and I hope you enjoy it!

Docs | Feed Feedback and Personalization Graze's Feed Feedback system uses Bluesky's "See More/See Less" buttons to create intelligent, personalized feed experiences that adapt to individual preferences while giving feed operators real-time ...

This helps in two ways:

① It will help me improve the feed (I see anonymized engagement reports).

② The feed will start to get customized to what you want to see more of and what you want to see less of.

A screenshot of a post highlighting the Show more like this, Show more less like this reporting feature.

You can help improve the feed!

If you see posts that don’t fit with the theme, or things you particularly want to see more or less of. Please follow these steps.

① Click the three dots in the bottom right corner of the post.
② Select the Show more/less like this from the menu.

For You, For Everyone - Graze Newsletter How Graze built a "composable personalization" engine for the open social web — and why it matters right now.

The personalization algorithm searches for posts that “people like you also liked” and generates a tailored feed accordingly.

For more details, read up on Graze's post about it:

This feed is run on a personalization algorithm based on what you have liked, and what similar users to you have liked, while it scans posts from the ML Ranked Feed list.

Think of it as a version of Spacecowboy's For You Feed, but for AI/ML.

A screenshot of the ML Feed: For You details block. Highlighting the Like and Pin buttons.

Welcome to ML Feed: For You!
A couple of brief details:

① If you are enjoying it, please pin it and give it a like

② Share and repost the feed so others might can find it

③ This feed is whitelisted, so while it looks for certain words and hashtags, they aren't required.

📌 ML Feed: For You • PINNED POST
(Expand this thread for feed info)

As genAI usage increases, health data exposure concerns rise: report | TechTarget Healthcare organizations are increasingly concerned about shadow AI and data protection as generative AI becomes part of clinical and administrative workflows.

Healthcare is shifting from "shadow AI" to secure, managed tools, with personal app usage dropping from 82% to 32%. Yet data risks remain high, regulated data accounts for 89% of policy violations.
#MedSky #MedAI

How Google and Taiwan are building an AI blueprint for public health Working with Google, Taiwan uses 20 years of health data and Gemini to bring predictive diabetes care to millions in its population-wide health system.

Google and Taiwan are proving AI can handle public health at scale. Their new model cuts diabetes risk assessment from 20 minutes to 25 seconds (a 14,400x efficiency boost). It’s a proactive blueprint for using 20 years of data to save clinician time and lives.
#MedSky 🛟

Large genome model: Open source AI trained on trillions of bases System can identify genes, regulatory sequences, splice sites, and more.

Evo 2 is here, and the scale is wild. It’s trained on nearly 9 trillion bases across all domains of life, picking out regulatory DNA and splice sites that often trip humans up. It’s fully open-source and already identifying critical BRCA2 mutations with impressive accuracy.
🧬💻

Polymarket Pulls Bet on Nuclear Detonation in 2026 ‘How ghoulish.’ The depravity economy moves into the nuclear war business.

Ah, so there is a bottom to the barrel.

Polymarket Pulls Bet on Nuclear Detonation in 2026

Automated detection of new cerebral infarctions and prognostic implications using deep learning on serial MRI - npj Digital Medicine npj Digital Medicine - Automated detection of new cerebral infarctions and prognostic implications using deep learning on serial MRI

Silent brain infarctions are easy to miss on serial MRI, but they're a huge warning sign. A new model with an 0.89 AUC found that "silent" lesions are linked to a 3.8-fold increased risk of a future stroke and could be a better way to manage risk.
#MedSky #MedAI

Radiology AI makes consistent diagnoses using 3D images from different health centres Vision–language model outperforms second-best models by an average of 20% across hospital sites.

Merlin is a new 3D AI model for abdominal CTs that’s actually holding up across different hospitals. Usually, these models tank when they see data from a new site, but Merlin averaged a 19.7% jump in accuracy over the next best system across three external centers.
#MedSky #MedAI

What AI Models for War Actually Look Like While companies like Anthropic debate limits on military uses of AI, Smack Technologies is training models to plan battlefield operations.

What AI Models for War Actually Look Like

www.wired.com/story/ai-mod...