Paul Kedrosky
speaker
1,259 appearances
4 recordings
2 series
first heard Apr 2026
last heard 6d ago
Paul Kedrosky’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 4 in all, peaking in Sep 2026 with 1.
Appearances
Showing how models- What do we do next?
Right.
Models in general, in a medical context, and this is writ large applies to models use in all complex environments.
They tend to over triage trivial cases, meaning that if you come in with like a cut, they're like, dude, this could be sepsis.
Let's take you in and start doing tissue biopsies.
And it's
It's like, no, no, it's just a cut.
Leave me alone.
And at the other end of the extreme, a woman comes in with chest, well, with pain in her back, which sometimes is indicative of some kind of cardiac event.
They're like, yeah, it's probably just a strain.
And so this idea of marching straight through and saying that the only thing that matters is the input data and therefore I can use these things in these complex environments.
We know these tendencies towards over triaging things.
trivial cases and under triaging critical ones.
That's also true.
Just as a side note, I gave a talk about this recently to the Fed where I was showing how models do the exact same thing in financial markets where they tend to become overaggressive when they should be conservative and vice versa, which leads to much more fragility in financial markets.
And yet, you know, we we march on.
And this is the deep problem is this kind of
complete misunderstanding of the nature of how these things respond in these complex environments and then the systemic consequences of doing it.
Like, for example, you replace radiologists with something with a tendency to over-triage, guess what you're going to get?
Far more testing, much more testing getting done, which may or may not be profitable for hospitals, but will have cascading consequences for people who have to have follow-up biopsies because of things that look like possibly malignancies and turns out they weren't.
Showing 761–780 of 1,259 · page 39 of 63
← Previous
Next →