Explore Wikipedia through a data map. Pages are grouped by semantic similarity, for topic clusters. Hover to see details, zoom to explore more fine-grained topics, click to go to a page. Search by page name to find interesting starting points for exploration. lmcinnes.github.io/datamapplot_...

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?

I updated my website! It is now just html and css. Possible blog post coming of the experience of updating it. Lots of fun older blog posts and newer projects are highlighted. colindcarroll.com

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.

Playing with integers: 2^1491 is the smallest power of 2 with 5 contiguous 0's. 2^1492 is the first with 6. This note is too narrow to contain the numbers (they have 449 and 450 characters)

Everything's crazy, so I implemented fractals in @matplotlib.bsky.social. Getting the transform API right, and figuring out how to repeatedly apply a function in Python were the hard parts. colab here if you want to try it out: colab.research.google.com/drive/10mzOt...

Time for us all to start studying! This is super well written, and seems super useful.

I used to have a bot that would reply to this whenever a different major paper used the word first. Might be time to resurrect it! (It was booted for being a bot 🤷)

git commit -m “No way I am gonna remember I accidentally used a double quote by the time I get to the end of this message’

Excited to share our new Nature paper on probabilistic weather forecasting today @neuripsconf.bsky.social! I'll be presenting at the DeepMind booth 1pm Vancouver time. I'll be at NeurIPS the whole week. If you're interested in AI for Science and sustainability, feel free to reach out for a chat!

Love it! This was my 2023 data, and I guess I got into bike commuting last year...

Could be a very good week.

I don’t have Spotify, but I use Strava. So I created my personal Year in Sports Wrapped 🚴 📊 Got Strava and want your own? No problem, I turned it into an app year-in-sports.streamlit.app Made in #python using #streamlit + #matplotlib, full code: github.com/Lisa-Ho/year...

If you're interested in a PhD at the intersection of machine learning and programming languages, consider applying to Yale CS! We're exploring new approaches to building software that draws inferences and makes predictions. See alexlew.net for details & apply at gsas.yale.edu/admissions/ by Dec. 15

There's a lot of room for some incredible incremental improvements using LLMs today! Give it a try! I get frustrated hearing claims about llms solving every problem (soon!), and people pointing out they don't (now!).

Introducing 🧞Genie 2 🧞 - our most capable large-scale foundation world model, which can generate a diverse array of consistent worlds, playable for up to a minute. We believe Genie 2 could unlock the next wave of capabilities for embodied agents 🧠.

thinking of calling this "The Illusion Illusion" (more examples below)

Proud I could do a straightedge and compass equilateral triangle without looking it up.

And here is where I would be since mid-2012! Interestingly, the answer is "about 20miles southwest" of where I started. The southernmost bit was the 2019 Boston Marathon. The two red lines in the west were the 2016 Napa Valley Marathon, and a run from Manchester, NH to Andover, MA.

Ever wonder where you would end up if you would end up if you started every run where you stopped the last one? I plotted that for my year. Turns out I'd be about 20 miles southeast of my house. Cool that I did the Pemi loop when I was at the most northern point of this "journey"!

D&D Combinatorics xkcd.com/3015

Similarly, when you try to convert your text into smaller pieces but all it does is tell you to throw yourself in next time, that's a fool of a tookenizer.

Allen is a great example of "what are the chances?" never being a rhetorical question.

Hello BlueSky! This is the official account for the ArviZ project. We provide FOSS tools for exploratory analysis of Bayesian models in both Python and Julia.

Another great seminar today is the Monte Carlo Online Seminar: sites.google.com/view/monte-c... Today's talk is on diffusion models!

Have a box that actually gets some use. The wood dowel hinge finally gave way. First attempt at a fix failed, but have great hope for attempt #2 (a real hinge).

New feature I'm trying to add is the ability to plot every run you've done, organized by distance. Cool to be able to spot the 1 mile track races at the top, and the Pemi loops at the bottom...

Updating github.com/ColCarroll/s... a little early this year (added some basic CI, typing, ruff, and the smallest tests possible). Here's my year so far... h/t to @matplotlib.bsky.social for this thing working for 6 years with no major changes!

What a chart. www.economist.com/interactive/...

Wanted to share that Varun Godbole recently released a prompting playbook. The title says prompt tuning, but this is text prompts, not soft prompts. github.com/varungodbole...

I recently got to work on a fun chapter, and the preprint is now available: arxiv.org/abs/2411.04260 We provide intuitions, examples, and recommendations for running MCMC in a world where GPUs and autodiff exist (like this world!)

#30DayMapChallenge Day 2: Lines Thought I'd try experimenting with terrain hachures. Wanted to capture multiple scales of surface form, which conventional hachures are not well suited to. As ever, data, source code and design notes available here: observablehq.com/@jwolondon/t...

Was confused about the construction, so wrote some code, and made some fun images, mostly by pushing the cardiod through a parametric function (instead of a circle, use a lemniscate or trifolium). If I was being careful, I suppose I'd use an arc-length parameterization, but I'm not being careful!

The only implementation I've seen of this (high quality!) sampler:

Boo! livefreeordichotomize.com/posts/2024-1...

Pete says "hey".

What makes an MCMC sampler GPU-friendly? statmodeling.stat.columbia.edu/2024/10/29/w...

New starter pack! go.bsky.app/GZ4hZzu

About 3,000 runs over 10 years, organized by distance.

Why did it take months to build this? I can't pay my rent with cargo-semver-checks. I appreciate all donations (♥!), but with a couple of exceptions (♥♥♥!) they've been $1-$5. It adds up to "monthly family dinner" not "rent." Please help me work on it more! github.com/sponsors/obi...

Used to be color when we had built-ins! Too hard to do on smaller shelves/cases.

The future of HMC: Multiscale Generalized Hamiltonian Monte Carlo with Delayed Rejection. Great talk by Bob Carpenter: www.youtube.com/watch?v=p32z... Basic idea: Run HMC, when you get a rejection, run it again but reduce the step-size. That way you can sample from the funnel of hell.

Joinery! I don't know why photos don't upload in any particular order. Just three more to go!

Still no stats, but started a pair of beefy step stools today. Made out of some cherry and mahogany boards I've had for too long. Going to try some fancy joinery on the top...

I promise I'm still doing statistics with computers, and hope to post more about that in the future, but in the interest of creating Content, here's a canoe paddle I made this summer out of a cherry board. It took about a month of 30-minutes-at-a-time, and was a lot of fun!

Parent: Ask your kid what it is! Other Parent: Never ask a child a question when you don’t know the answer Me: I think that’s actually cross-examination and not kids?