Robin Blythe
@rbly.bsky.social
Assistant professor at Duke-NUS medical school. Mostly interested in health economics, biostats and clinical informatics.
A colleague (who prefaced this with "don't repeat this, but") said that treating kids is like being a vet but worse, because they can't articulate anything, just make noises, and you can't knock them out. Hence, most specialists stay as far away as possible.
Sad, but also, lol
Sad, but also, lol
February 5, 2026 at 9:46 AM
A colleague (who prefaced this with "don't repeat this, but") said that treating kids is like being a vet but worse, because they can't articulate anything, just make noises, and you can't knock them out. Hence, most specialists stay as far away as possible.
Sad, but also, lol
Sad, but also, lol
There have been some attempted solutions to this phenomenon in the past, like health system navigators or assigned social workers. They're usually critically insufficient or just flat out ineffective, from what I remember. I wonder whether we need to treat more acutely.
February 5, 2026 at 9:17 AM
There have been some attempted solutions to this phenomenon in the past, like health system navigators or assigned social workers. They're usually critically insufficient or just flat out ineffective, from what I remember. I wonder whether we need to treat more acutely.
What's nuts is that this happens across many facets of healthcare. In dentistry there is apparently a pretty well-known phenomenon of more experienced clinicians picking their long-term, easy clients and dumping the complex patients on trainees or even students.
February 5, 2026 at 9:13 AM
What's nuts is that this happens across many facets of healthcare. In dentistry there is apparently a pretty well-known phenomenon of more experienced clinicians picking their long-term, easy clients and dumping the complex patients on trainees or even students.
p(o, k|\(\theta \)) = 0.8
February 5, 2026 at 8:33 AM
p(o, k|\(\theta \)) = 0.8
This is probably less intuitive for the non-Aussies but the joke is that the params look like "r u o k"
February 5, 2026 at 7:44 AM
This is probably less intuitive for the non-Aussies but the joke is that the params look like "r u o k"
Phew, glad you weren't meant to eat it as a burrito
February 5, 2026 at 7:22 AM
Phew, glad you weren't meant to eat it as a burrito
Is it in a flour tortilla?
February 5, 2026 at 12:46 AM
Is it in a flour tortilla?
"Multiverse analysis: Modelling for people with commitment issues"
February 4, 2026 at 5:04 AM
"Multiverse analysis: Modelling for people with commitment issues"
I think they are proprietary, sadly - will ask though! In all honesty it won't be anything you don't already know, it was very introductory
February 4, 2026 at 4:59 AM
I think they are proprietary, sadly - will ask though! In all honesty it won't be anything you don't already know, it was very introductory
In some sense it's a little like thinking about how you'd provide healthcare for a critically underresourced LMIC...
February 4, 2026 at 4:58 AM
In some sense it's a little like thinking about how you'd provide healthcare for a critically underresourced LMIC...
It's so vapid, made worse by vapid people covering it during the pandemic lol
February 4, 2026 at 2:21 AM
It's so vapid, made worse by vapid people covering it during the pandemic lol
This is a core problem of LLMs generally, I think. Ultimately I wouldn't be comfortable letting them too near prediction, the certainty overwhelms decisions unless there's strong prior knowledge
February 3, 2026 at 11:44 PM
This is a core problem of LLMs generally, I think. Ultimately I wouldn't be comfortable letting them too near prediction, the certainty overwhelms decisions unless there's strong prior knowledge
I haven't come across these before, thanks for pointing them out! Seems like the key assumption here is that the residuals reflect all types of error, including poor model fits, rather than just noise.
February 2, 2026 at 2:47 PM
I haven't come across these before, thanks for pointing them out! Seems like the key assumption here is that the residuals reflect all types of error, including poor model fits, rather than just noise.
Prediction modelling? Admittedly, stretching the definition of analysis or frequentism, though.
February 2, 2026 at 5:10 AM
Prediction modelling? Admittedly, stretching the definition of analysis or frequentism, though.
Relevant (and excellent) XKCD
youtu.be/WYf9-xfm6t8?...
youtu.be/WYf9-xfm6t8?...
February 2, 2026 at 5:05 AM
Relevant (and excellent) XKCD
youtu.be/WYf9-xfm6t8?...
youtu.be/WYf9-xfm6t8?...
Feel like you learn much about Econ? Or is that not the point?
February 2, 2026 at 2:02 AM
Feel like you learn much about Econ? Or is that not the point?
Wow, it's been going since 2006. I've never actually listened to it, should probably give it a go.
February 2, 2026 at 12:30 AM
Wow, it's been going since 2006. I've never actually listened to it, should probably give it a go.
Does that say 18 years ago?
February 2, 2026 at 12:28 AM
Does that say 18 years ago?
The reality is usually somewhere in the middle. But yes I think it lines up closer to Gigerenzer in this case.
February 2, 2026 at 12:27 AM
The reality is usually somewhere in the middle. But yes I think it lines up closer to Gigerenzer in this case.
Going to start calling the Ashes "The Superbowl, because the bowl the ball" to trigger my Aussie friends
February 1, 2026 at 11:58 AM
Going to start calling the Ashes "The Superbowl, because the bowl the ball" to trigger my Aussie friends
Please tell me this is going in an academic paper so I can cite it
February 1, 2026 at 11:57 AM
Please tell me this is going in an academic paper so I can cite it