Speaking of mistakes, you'll note that although we're supposed to deliver on Thursdays, I appear to have sent this week's post a day earlier than intended. Oops!
Why was reaction time increasing before intervention? Can you get the LLMs to find the changepoints ( drastic changes in the rolling mean) without any data besides performance and date? If they end up right on or just after the vertical bars separating the interventions that would be more evidence. I also want to know if they are capable of using the technique.
> Why was reaction time increasing before intervention?
Good question. I'm not sure. I think I may not have been taking the test as often as I had been earlier in the baseline period, and the reduced practice may have made me slower. Note that there was a cluster of more-frequent tests during the baseline period. That cluster corresponds to a period of time when I was doing the test three times each day as part of my earlier creatine experiment (which showed, at best, a minimal effect of creatine supplementation).
I have had similar experiences on quantified mind.
From the plot it seems 500 mg dose reduced reaction time over time but 1000 mg immediately gave the entire advantage metformin could provide. I have never heard of that before though it does make sense.
I'm not sure, but I think if I had continued with the 500 mg dose, I would have plateaued at a level fairly close to what I achieved with 1000 mg. My hunch is based on just eyeballing the plot -- not on any kind of analysis.
I could not get Chatgpt 5.1 pro to find changepoints despite asking it to use various techniques and manipulating the data several ways. Did make several 'non-parametric' plots though. I don't know how to post them here though. I could share the pro power by adding people to group chat?
Speaking of mistakes, you'll note that although we're supposed to deliver on Thursdays, I appear to have sent this week's post a day earlier than intended. Oops!
I think you were perhaps more focused on the idea of expeditiously making a correction, which is the mark of a good scientist.
Damn. And there was me thinking that tomorrow would be Friday.
Richard, thanks for the nice write-up.
Why was reaction time increasing before intervention? Can you get the LLMs to find the changepoints ( drastic changes in the rolling mean) without any data besides performance and date? If they end up right on or just after the vertical bars separating the interventions that would be more evidence. I also want to know if they are capable of using the technique.
> Why was reaction time increasing before intervention?
Good question. I'm not sure. I think I may not have been taking the test as often as I had been earlier in the baseline period, and the reduced practice may have made me slower. Note that there was a cluster of more-frequent tests during the baseline period. That cluster corresponds to a period of time when I was doing the test three times each day as part of my earlier creatine experiment (which showed, at best, a minimal effect of creatine supplementation).
I have had similar experiences on quantified mind.
From the plot it seems 500 mg dose reduced reaction time over time but 1000 mg immediately gave the entire advantage metformin could provide. I have never heard of that before though it does make sense.
I'm not sure, but I think if I had continued with the 500 mg dose, I would have plateaued at a level fairly close to what I achieved with 1000 mg. My hunch is based on just eyeballing the plot -- not on any kind of analysis.
I could not get Chatgpt 5.1 pro to find changepoints despite asking it to use various techniques and manipulating the data several ways. Did make several 'non-parametric' plots though. I don't know how to post them here though. I could share the pro power by adding people to group chat?
IT WORKED ! Just had to turn on deep science think mode. https://chatgpt.com/s/t_69313b3adb848191b93f630484a0084a still dont know how to share the pictures.