jSoboil
@iamaloop.bsky.social
I like to saunter and paint, but I seem to have a real shtick for Health Economics | Primum non nocere | @rhta16 | @rhta_lmics
Reposted by jSoboil
I occasionally delude myself that I know something about stats. Today I learned a word from my 6 year old granddaughter: “subitising”. I told her she must mean “summarising”. Now that she’s asleep I have looked it up. Apologies will be due at breakfast.
April 17, 2025 at 6:39 PM
I occasionally delude myself that I know something about stats. Today I learned a word from my 6 year old granddaughter: “subitising”. I told her she must mean “summarising”. Now that she’s asleep I have looked it up. Apologies will be due at breakfast.
Reposted by jSoboil
I've never experienced anything like this in my life, where people keep telling me that "we have to use AI or else we'll be left behind" but they don't actually tell me exactly what problem they are hoping to solve with the "AI" or how we would know it was useful. But we must use it! Apparently.
March 19, 2025 at 4:11 PM
I've never experienced anything like this in my life, where people keep telling me that "we have to use AI or else we'll be left behind" but they don't actually tell me exactly what problem they are hoping to solve with the "AI" or how we would know it was useful. But we must use it! Apparently.
Reposted by jSoboil
Of course they do. Statistical models can work this way too. The algorithms, in this sense, are not the problem. The problems lie in the foolish ways some try to apply them, and a broad lack of interest in actually evaluating whether the predictions they generate are clinically useful.
The authors point out that AI models base their predictions on sneaky shortcut effects all the time; they're just easier to identify when the conclusions (beer drinking) are clearly spurious.
www.nature.com/articles/s41...
www.nature.com/articles/s41...
The risk of shortcutting in deep learning algorithms for medical imaging research - Scientific Reports
Scientific Reports - The risk of shortcutting in deep learning algorithms for medical imaging research
www.nature.com
December 16, 2024 at 12:15 PM
Of course they do. Statistical models can work this way too. The algorithms, in this sense, are not the problem. The problems lie in the foolish ways some try to apply them, and a broad lack of interest in actually evaluating whether the predictions they generate are clinically useful.
Reposted by jSoboil
That's absolutely true. It's an issue that students with a math background can struggle with because they treat the model as given, rather than something to be tested against data and refined.
December 9, 2024 at 8:36 AM
That's absolutely true. It's an issue that students with a math background can struggle with because they treat the model as given, rather than something to be tested against data and refined.
Reposted by jSoboil
When clinical arguments compete against statistical arguments and the clinical arguments are valid, we can't have nice things if the statistical arguments override reason.
(The paper is nice, but arguing that one should listen to not well justified arguement on the basis of MSE is erroneous)
(The paper is nice, but arguing that one should listen to not well justified arguement on the basis of MSE is erroneous)
November 19, 2024 at 10:49 PM
When clinical arguments compete against statistical arguments and the clinical arguments are valid, we can't have nice things if the statistical arguments override reason.
(The paper is nice, but arguing that one should listen to not well justified arguement on the basis of MSE is erroneous)
(The paper is nice, but arguing that one should listen to not well justified arguement on the basis of MSE is erroneous)