Colin Carroll
banner
colcarroll.bsky.social
Colin Carroll
@colcarroll.bsky.social
Runner, biker, hiker. Software engineer @DeepMind, and open source enthusiast. Sometimes crafts things out of wood. he/his.
I too just upgraded. So far: it is fine! I feel like I could probably have pushed my pixel 7 for another year or two though.
September 3, 2025 at 10:14 PM
Chad! This is a very confused ❤️.
May 28, 2025 at 10:15 PM
I was able to try updating my site recently. I (finished! then) wrote it up but the conclusion was similar that getting started was easy, but it took the same total time: bsky.app/profile/colc...
Ok, first new blog post in 6 years, about updating my website with the help of an LLM: colindcarroll.com/blog/llm_blo...

maybe I will write about statistics again next?
Updating my website with an LLM
colindcarroll.com
May 16, 2025 at 10:07 PM
Jinx!
May 13, 2025 at 12:22 AM
Page not found!
May 13, 2025 at 12:22 AM
I have no heuristics for 19 minutes (except more than 1000 seconds) *or* 19mph (except faster than I can run).

Prime numbers are the work of the devil.
May 6, 2025 at 5:52 PM
College sports (5k and 10k on the track!) prepared me for the real world by letting me know 1,000 seconds is 16:40.

If it takes you 16:40 to run a 5k, that's ~5:20/mi, or exactly 5m/s.

Surprisingly useful for mental math, like when something (presumably stan) takes 168 minutes.
Chain 2 finished in 10083.8 seconds.
May 6, 2025 at 2:54 PM
...I think I just heard it go by.
April 4, 2025 at 6:04 PM
Should I be watching for a flyover?
April 4, 2025 at 6:01 PM
A naive Python program can do this search up to 8 contiguous zeros (2^14007) in about a minute. That's 4217 digits long: 4kb as a string!

I wasn't able to find 9 contiguous zeros, letting the program run for 10 minutes, up to 2^70000.
March 20, 2025 at 9:25 PM
I guess it makes sense: if a number has n zeros in a row, twice that number has at least n-1?
March 20, 2025 at 9:16 PM
Get some sleep, you wacky plotting library!
February 13, 2025 at 2:30 AM
But for multimodal densities, note that the usual adaptation methods can be actively harmful!
February 2, 2025 at 2:03 PM
We mostly focus on scaling out to lots of chains or huge log likelihoods. Some of the ensemble chain adaptation methods, and some intuitions for designing algorithms for accelerators might be useful to think about.
February 2, 2025 at 2:02 PM
What, print it out on a flat sheet of paper? Who is that going to impress?
December 29, 2024 at 3:34 AM