I know there are a lot of people working on generative AI and foundational models in OHDSI, and I think it would be good to at least be aware of each other and what we’re doing. This is why I’m launching the Generative AI and Analytics in Healthcare (GAIA) Workgroup.
To be clear, this includes both research around applying existing generative language models such as ChatGPT, as well as training new foundational models, either on textual data or structured data such as the data in the OMOP CDM. I thought about splitting these up, but I hope that at some point they will converge.
If you’re interested, please visit the Team I set up in our OHDSI Teams environment.
I’m actually looking to collaborate on projects that involves real world healthcare Data…
Anyone doing any machine learning/artificial intelligence project in any area relating to that could contact me, I would render assistance to the best of my ability… I’m a pharmacist and data scientist.
Great minds think alike? Or more likely we’re all just running out of Greek mythology names at this point.
The real question is if rightful name ownership is determined by strength of conceptual alignment to the deity, or instead by strength and flow of the acronym? Maybe it says in the Book of OHDSI somewhere.
I hope I speak for @Andrew and @kzollove when I say - we are prepared to barter and have had our eyes on the name ATLAS for awhile.
I have made the executive decision to go for the second Tuesday of each month, at noon Eastern Time.
However, I can’t make the second Tuesday of February, so, just to maximize chaos I also decided we’ll have our kickoff meeting on the first Tuesday of February, at noon Eastern Time.
So, please all join on February 6, at noon Eastern Time. Here’s the meeting link for the kickoff meeting. If you joined the Team I think you should have also gotten an invite by mail.
This Tuesday (March 12) we’ll have our second Generative AI and Foundational Models Workgroup meeting. On the agenda is a presentation by Michael Wornow, first author of The shaky foundations of large language models and foundation models for electronic health records. Please join! The meeting starts at noon Eastern Time, which is 5pm Central European Time (because the US just went of DST, but Europe didn’t)
Tomorrow (Tuesday April 9) we have our next workgroup meeting. This meeting I’d like to discuss sharing of models, code, and evaluation benchmarks for Foundational Models for Electronic Health Records (FMEHRs). One specific topic of discussion will be Stanford’s MEDS standard for data as input for FMEHRs.
(FMEHRs are models pre-trained on all data in the CDM, not models pre-trained on all text on the internet like GPT-4)
The meeting will be at
6pm Central European Time
Noon Eastern Time
9am Pacific Time
The meeting will be recorded for those in other time zones.
Tomorrow (June 11) we have our next workgroup meeting. This meeting I’d like to focus on building (open) APIs for AI agents; We want LLMs such as ChatGPT to be aware of information generated in OHDSI, such as
The content of the vocabulary
The structure and conventions of the CDM
The databases available for network studies and their characteristics
The evidence generated by OHDSI, for example all characterizations and causal effect estimates generated in LEGEND studies.
For this, we’ll need to create APIs that AI agents can call on. We’ll have presentations by João Almeida and Hayden Spencer on their work in this space, after which we can brainstorm on further OHDSI activities.
The meeting will be at
6pm Central European Time
Noon Eastern Time
9am Pacific Time
The meeting will be recorded for those in other time zones.
