Adam Rodman
@AdamRodmanMD
Followers
18K
Following
15K
Statuses
15K
Physician, educator, historian, author, podcaster, researcher @BIDMC_IM @HarvardMed, host of #histmed podcast @BedsideRounds, AE @NEJM_AI, studies 🤖+🧠. 🖖🚲
Boston, MA
Joined March 2010
RT @OldSchoolBoston: At Dusk (Boston Common at Twilight) by Frederick Childe Hassam, 1885-86.
0
52
0
@raj_mehta An interesting perspective! So like NLP extraction of symptoms to identify new nosologies that we wouldn't have thought of otherwise?
1
0
2
@iandarin Yes! What you're getting at is the idea of a machine nosology, which would be the counter force to this (ie linking of symptoms/findings directly to outcomes using large data sets that can be shown to empirically work, but isn't understandable by us)
1
0
0
@CoryRohlfsen @DavidDeutschOxf @ToKTeacher @perigean @techczech IMO that's 100% the case, especially when you take a longer view at the multiple nosologies we've had in human history
0
0
1
@reverendofdoubt I don't understand it either 🤣 I'm an accidental researcher... Still a clinician/educator at heart!
0
0
2
@reverendofdoubt (suspect your research is like mine, with little overhead other than compute and personnel)
1
0
1
@perigean @CoryRohlfsen @DavidDeutschOxf @ToKTeacher @techczech Asking because I'm genuinely curious -- can you explain more?
1
0
0
@CoryRohlfsen @DavidDeutschOxf @ToKTeacher @perigean @techczech And your question is good -- is it knowledge? Is knowledge just a human construct? Or is it something else altogether?
1
0
0
@hortonhearsafoo @RichardAFerraro WebUI? That's what we are doing at HMS! (And what a lot of academic institutions have done). Given improvements in open source models (even the 70b ones) at some point I assume some places will have clusters we can use for real time inference!
1
0
2
@doc_BLocke @phlegmfighter The pulm nodule case was one of the dramatic LLM improvements. Much more nuanced thinking about patient prefs and uncertainty in work up (W/o LLM most ppl just said, transitional issue)
1
0
3
@RichardAFerraro For our study we used research accounts of ChatGPT, cases had no no PHI (and were quite modified through our Delphi process anyway)
0
0
0