[Chen Samulsion]

Almost every progress bar on your computer is a fabrication. It is not measuring data transfer. It is managing your anxiety. The history of why we built an entire discipline around making computers feel slower than they are.

https://ux.stackexchange.com/questions/11881/progress-bars-why-are-they-never-useful

https://rootbyte.tech/article/progre...sychology-1985

[Chen Samulsion]

Can we have a pseudo twoway bar command that runs like twoway line? That will make graph easier just like below. Name the pseudo command twoway bars... (The following graph was made by MS Excel. Data is sysuse sp500.)

[Erik Ruzek]

Coming back to #174, Nigel Moore Claude Code was able to rapidly port those tests from the R package PMCMRplus to Stata+Mata code, create a validation report, and a help file (<10 min). Sometimes when you port a code file, it doesn't validate the method, only the port so we need to be careful. This is why the original source materials and human domain expertise is important. I have found that Claude is quite good at going from R code to Stata code, but relatively less good (but not bad) at de novo coding solutions in Stata vs R.

I agree with this but also the models are improving rapidly. So any statements we make about what they are good at right now will be stale in 6 months, and moldy in a year. Stata/Mata is simply less used than Python, R, C/C++, and the like. Accordingly, the models are at a disadvantage when doing something where there is no reference in another software language they have more training data on.