Scott McGrath
@smcgrath.phd
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/
https://citris-uc.org/people/person/scott-mcgrath/
A second lesson in AI governance (To Epic or Epic): When AI tools are inexpensive to deploy and maintain, prioritization is driven by internal resourcing needs. This often results in more leniency, allowing mild-moderately effective tools to persist.
#AMIA2025 #MedSky #MLSky
#AMIA2025 #MedSky #MLSky
November 17, 2025 at 3:46 PM
Smitha Ganeshan shares that UCSF Health utilizes a two-tiered AI governance model. The "Health AI Steering" committee handles prioritization and resourcing, while the "Health AI Oversight" committee guides the evaluation and responsible deployment of ethical AI.
#AMIA2025 #MedSky #MLSky
#AMIA2025 #MedSky #MLSky
November 17, 2025 at 3:46 PM
Ysair emphasized then need to educate leadership and clinicians how harmful some of the AI "bad apples" can be. He had a proud Dad moment when his 8-year old said "ChatGPT is kind of stupid".
#AMIA2025
#AMIA2025
November 17, 2025 at 3:31 PM
Ysair emphasized then need to educate leadership and clinicians how harmful some of the AI "bad apples" can be. He had a proud Dad moment when his 8-year old said "ChatGPT is kind of stupid".
#AMIA2025
#AMIA2025
Yasir Tarabichi from MetroHealth approaches AI deployment as an implementation science discipline, emphasizing that governance and validation cannot be outsourced. They are fast-following mature tools while co-developing new, integrated EHR solutions.
#AMIA2025 #MedSky #MLSky
#AMIA2025 #MedSky #MLSky
November 17, 2025 at 3:31 PM
Yiye Zhang from Weill Cornell Medicine talks about research on using LLMs to predict ED revisits leading to admission found a significant challenge.
A model showing moderate performance at WCM showed noticeable deterioration at VUMC, highlighting generalization issues.
#AMIA2025 #MedSky #MLSky
A model showing moderate performance at WCM showed noticeable deterioration at VUMC, highlighting generalization issues.
#AMIA2025 #MedSky #MLSky
November 17, 2025 at 3:20 PM
Brian Patterson from UW Med School talks about the landscape for AI adoption is shifting. Pre-2025 "by the use" AI pricing prompted ROI discussions. Post-2025, tools are increasingly "bundled" into platforms as vendor releases & clinician demand grow, highlighting safety and rigor. #AMIA2025 #MedSky
November 17, 2025 at 3:11 PM
Brian Patterson from UW Med School talks about the landscape for AI adoption is shifting. Pre-2025 "by the use" AI pricing prompted ROI discussions. Post-2025, tools are increasingly "bundled" into platforms as vendor releases & clinician demand grow, highlighting safety and rigor. #AMIA2025 #MedSky
@kdpsingh.bsky.social focusing on the "build before you buy" mentality. A lot of the AI tools that are most useful to clinicians aren't actually commercially available, so internal teams build tools they need, like their NoteReader which can help automate measurements.
#AMIA2025 #MLSky #MedSky
#AMIA2025 #MLSky #MedSky
November 17, 2025 at 3:00 PM
@kdpsingh.bsky.social focusing on the "build before you buy" mentality. A lot of the AI tools that are most useful to clinicians aren't actually commercially available, so internal teams build tools they need, like their NoteReader which can help automate measurements.
#AMIA2025 #MLSky #MedSky
#AMIA2025 #MLSky #MedSky
GenAI presents both opportunities and challenges for educators, @billhersh.bsky.social shares his experiences, referencing @emollick.bsky.social who has said we face a "homework apocalypse"
#AMIA2025 #AcademicSky #MLSky #MedSky
#AMIA2025 #AcademicSky #MLSky #MedSky
November 17, 2025 at 2:10 PM
GenAI presents both opportunities and challenges for educators, @billhersh.bsky.social shares his experiences, referencing @emollick.bsky.social who has said we face a "homework apocalypse"
#AMIA2025 #AcademicSky #MLSky #MedSky
#AMIA2025 #AcademicSky #MLSky #MedSky
Jessica S. Ancker from Vanderbilt Medical Center presented on collaboration and transparency in using AI.
#AMIA2025 #AcademicSky
#AMIA2025 #AcademicSky
November 17, 2025 at 2:05 PM
Jessica S. Ancker from Vanderbilt Medical Center presented on collaboration and transparency in using AI.
#AMIA2025 #AcademicSky
#AMIA2025 #AcademicSky
@insarkar.bsky.social from Brown University explored the practical integration of LLMs into informatics education, moving "From Prompt to Practice."
#AMIA2025 #Academicsky #MLSky
#AMIA2025 #Academicsky #MLSky
November 17, 2025 at 1:51 PM
@insarkar.bsky.social from Brown University explored the practical integration of LLMs into informatics education, moving "From Prompt to Practice."
#AMIA2025 #Academicsky #MLSky
#AMIA2025 #Academicsky #MLSky
How can educators prompt curiosity in their students' prompting?
Mark Weiner from Regeneron explored this meta-level educational challenge due to the era of LLMs.
#AMIA2025 #AcademicSky
Mark Weiner from Regeneron explored this meta-level educational challenge due to the era of LLMs.
#AMIA2025 #AcademicSky
November 17, 2025 at 1:47 PM
How can educators prompt curiosity in their students' prompting?
Mark Weiner from Regeneron explored this meta-level educational challenge due to the era of LLMs.
#AMIA2025 #AcademicSky
Mark Weiner from Regeneron explored this meta-level educational challenge due to the era of LLMs.
#AMIA2025 #AcademicSky
#AMIA2025 Symposium has 2,145 people here!
Of 1,977 submissions, overall acceptance rates included:
Panels at 37%,
Regular Papers at 43%,
and Posters at 61%.
A strong presence of student and physician coauthors was called out.
Of 1,977 submissions, overall acceptance rates included:
Panels at 37%,
Regular Papers at 43%,
and Posters at 61%.
A strong presence of student and physician coauthors was called out.
November 16, 2025 at 6:59 PM
#AMIA2025 Symposium has 2,145 people here!
Of 1,977 submissions, overall acceptance rates included:
Panels at 37%,
Regular Papers at 43%,
and Posters at 61%.
A strong presence of student and physician coauthors was called out.
Of 1,977 submissions, overall acceptance rates included:
Panels at 37%,
Regular Papers at 43%,
and Posters at 61%.
A strong presence of student and physician coauthors was called out.
November 16, 2025 at 6:46 PM
November 16, 2025 at 6:45 PM
For the next step of the workshop: The goal of data synthesis is to transform extracted data into meaningful insights. AI can accelerate this by suggesting themes, but humans provide the interpretive rigor: verifying relevance, interpreting meaning, and ensuring transparency.
#AMIA2025 #AcademicSky
#AMIA2025 #AcademicSky
November 16, 2025 at 3:36 PM
For the next step of the workshop: The goal of data synthesis is to transform extracted data into meaningful insights. AI can accelerate this by suggesting themes, but humans provide the interpretive rigor: verifying relevance, interpreting meaning, and ensuring transparency.
#AMIA2025 #AcademicSky
#AMIA2025 #AcademicSky
Continuing the workshop: AI can assist data extraction by organizing info into structured tables and batch-extracting fields from multiple sources.
Human reviewers remain essential. They must verify the extracted data for accuracy, interpret context, & resolve discrepancies.
#AMIA2025 #AcademicSky
Human reviewers remain essential. They must verify the extracted data for accuracy, interpret context, & resolve discrepancies.
#AMIA2025 #AcademicSky
November 16, 2025 at 2:55 PM
Continuing the workshop: AI can assist data extraction by organizing info into structured tables and batch-extracting fields from multiple sources.
Human reviewers remain essential. They must verify the extracted data for accuracy, interpret context, & resolve discrepancies.
#AMIA2025 #AcademicSky
Human reviewers remain essential. They must verify the extracted data for accuracy, interpret context, & resolve discrepancies.
#AMIA2025 #AcademicSky
Evidence synthesis is essential, but repetitive work. AI can assist by generating preliminary classifications and summarizing abstracts, but human reviewers must verify recommendations, interpret context, and resolve ambiguities. #AMIA2025 #AcademicSky
November 16, 2025 at 1:52 PM
Evidence synthesis is essential, but repetitive work. AI can assist by generating preliminary classifications and summarizing abstracts, but human reviewers must verify recommendations, interpret context, and resolve ambiguities. #AMIA2025 #AcademicSky
DAPT and Supervised Fine-Tuning serve different goals. DAPT uses unlabeled domain text, like PubMed, to teach a model medical language. SFT uses labeled instruction-response data to train a model for a specific task, such as creating a clinical summarizer.
#AMIA2025 #MedSky #MLSky
#AMIA2025 #MedSky #MLSky
November 15, 2025 at 6:48 PM
