Bjorn Watsjold

Emergency Physician, Medical Education Research(er?), possible polemicist. Opinions, at most, my own.

[Bjorn Watsjold @docbjorn]

@ivanzvonarmd at a Pels game you might get this

[Pelicans Film Room @PelsFilmRoom]

OH MY GOD WHAT A FINISH BY ZION

[Bjorn Watsjold @docbjorn]

@doc_BLocke @JAMAInternalMed Protecting the trope “the procedure was technically perfect, nonetheless the patient died.”

[Bjorn Watsjold @docbjorn]

@doc_BLocke @JAMAInternalMed Though in this case, (2) looks to be a logic statement, if something is adjudicated to be an error, there is a process fault, thus, if we can’t find a process fault, there isn’t an error

[Brooks Walsh @BrooksWalsh]

This report is a short but powerful review of police dog injuries, and the injustices underlying their use.
I want to illustrate their report with a few of the cases
🧵

[Bjorn Watsjold @docbjorn]

Supervising medical student sim lab—there should be acting awards for playing sim and standardized patients

[Bjorn Watsjold @docbjorn]

@jeremyfaust Also, hard disagree that we think like LLMs, but that may be a discussion for another day

[Bjorn Watsjold @docbjorn]

@jeremyfaust But as we engage further with “how to think with GPT,” move towards authorship criteria, validity or liability of judgments made with AI, we need better metaphors for how this works and what is happening. We can blame AI as a separate agency, but we’re behind the curtain

[Bjorn Watsjold @docbjorn]

@jeremyfaust A better metaphor might be that of Otto’s notebook, in which we consider whether a notepad used to record and recall ideas and plans is part of Otto’s thinking or is just an artifact. GPT can present statistical prose, soon will be able to calculate and may even get Bayesian…

[Bjorn Watsjold @docbjorn]

@jeremyfaust This seems to confound the statistical generation of prose with reasoning and contributes to the anthropomorphism of gpt. I think that’s a mistake, despite the useful ways you’ve applied it to your work.

[Bjorn Watsjold @docbjorn]

@EmpiricGame @dr_dmorgan @AdamRodmanMD @ldscherer Completely agree. I don’t know both specificity and sensitivity for any test off the top of my head, much less apply them in practice, but I also think few doctors have a sense of baseline/pretest risk. Nonetheless, improved testing numeracy helps conceptually when using tests.

[Bjorn Watsjold @docbjorn]

@EmpiricGame @dr_dmorgan @AdamRodmanMD @ldscherer This is strong evidence that Gerd Gigerenzer should be taught in medical school…would still be simpler if it was just likelihood ratios

[Attorney Alex Taubes @alextt]

Privatization of public services — hospitals, schools, prisons — results in lower quality services. Worse, it results in corruption, theft of public dollars, and fraud.

CT allowed a national healthcare company to buy three community hospitals—a disaster!
https://t.co/XVv0iy6voA

[Bjorn Watsjold @docbjorn]

@TChanMD @erikwdriessen @mededdoc There may still be room for a reflection on how validity arguments should be made for the use of LLMs

[Bjorn Watsjold @docbjorn]

Want to be one of the most cited researchers over the next 20 years? Develop STARD or similar reporting guidelines for AI-classified data. Assessments will be automated, how much bias will be hidden if we don't anticipate how and why it occurs? @mededdoc @erikwdriessen @TChanMD

[Bjorn Watsjold @docbjorn]

@TChanMD @erikwdriessen @mededdoc I should have spent 2 minutes on google scholar, @dmoher may have already done it https://t.co/qkaG81K7iH

[Bjorn Watsjold @docbjorn]

@TChanMD @erikwdriessen @mededdoc I feel like we need someone with a little more background in LLMs to be the insider? I’ve done a lot of skeptical reading, but we probably need a counterpoint

[Bjorn Watsjold @docbjorn]

@TChanMD @erikwdriessen @mededdoc The original tweet was more a request to the #meded world, but if a team is assembling we might just have to go ahead and do it.

[Bjorn Watsjold @docbjorn]

@EmpiricGame Gotcha, and heart player gets stuck with all the fluid cards even though their hand is too full

[Bjorn Watsjold @docbjorn]

@EmpiricGame Could also develop objectives that can’t be completed individually to reflect interdependence