iiinova — where your digital self keeps evolving

will add - the test file is the audio of the "I Have a Dream" speech, which is 6:46 and takes about 4 seconds to decode on my machine. I also added it to my tap, so `brew install llimllib/tap/transcribe` will install it

hyperfine benchmark showing the 6:46 audio from MLK's "I Have a Dream" speech translates in about 3.95 seconds on my machine

120

llimllib @billmill.org · 24/09/2026

I'm guessing that site is missing this one from my childhood

010

llimllib @billmill.org · 24/08/2026

Im in Oslo on a family vacation and got to play disc golf at one of the top courses in the world this morning. Played awful but had a lot of fun

160

llimllib @billmill.org · 04/08/2026

It had been 10 years since I did a "Solving Sol", and for some reason I got the idea to do one this morning. You can play around with it at llimllib.github.io/solving-sol/... if you're so inclined

111

llimllib @billmill.org · 12/07/2026

Went out to play disc golf and got attacked by a pair of Kingbirds. Merlin’s description is accurate

030

llimllib @billmill.org · 02/07/2026

Goals scored so far this world cup, by age of player

A plot of goals scored at the world cup by age. Messi and Ronaldo stand apart, as they have for so long now

100

llimllib @billmill.org · 21/06/2026

I took a photo of the sky over your area at 11:45 as we packed up to head home

111

llimllib @billmill.org · 16/06/2026

Iran has the oldest squad in the World Cup, Côte d'Ivoire the youngest

Joy Division plot of World Cup squad ages

1111

llimllib @billmill.org · 14/06/2026

Nobody asked me for a Joy Division plot of squad height, but you're getting it anyway

A Joy Division plot of squad height at the 2026 FIFA world cup

141

llimllib @billmill.org · 14/06/2026

My brother asked me about average world cup squad heights, so I could hardly help myself. Turns out that Netherlands vs. Japan is only the 7th largest height differential game we'll see in the group stages; Bosnia vs Qatar is the largest

110

llimllib @billmill.org · 12/06/2026

looks like one match at least is flipped?

110

llimllib @billmill.org · 10/06/2026

I ran a test of container file system performance on Mac OS by the simple expedient of measuring `npm install` with a cold cache. github.com/llimllib/con... Orbstack did shockingly well, paying basically no performance tax whatsoever for running in a container. I've not previously used it

a graph showing the results of a benchmark comparing `npm install` performance on a Mac with several container frameworks

110

llimllib @billmill.org · 09/06/2026

Leading causes of death among Children and Adolescents in the United States source: www.nejm.org/doi/full/10.... guns and cars are what kill kids

guns and cars kill about 2x kids as any other cause in the United States from 2018-2023, per New England Journal of Medicine

120

llimllib @billmill.org · 18/05/2026

Here it is per 36, no fair crediting the spurs for the extra minutes they've played

000

llimllib @billmill.org · 18/05/2026

Strictly for kicks, here's playoff net points vs height for the two teams:

a scatterplot showing net points vs height for the Thunder and Spurs. On the Y axis is net points so far in the playoffs, and on the X axis is height in inches

100

llimllib @billmill.org · 17/05/2026

with a remote control like this

380

llimllib @billmill.org · 14/05/2026

Victor has a pretty bimodal distribution, just that his "bad" games are -0.5

020

llimllib @billmill.org · 02/05/2026

And the championship season

Boston Celtics players net points, 2023-24 playoffs

000

llimllib @billmill.org · 02/05/2026

For comparison, here's last year's performances: (I added a dodge so the circles don't overlap. I wish plot supported a dodge that overlapped, but afaict it doesn't)

Celtics players performances by net points in last year's postseason

100

llimllib @billmill.org · 02/05/2026

Raptors vs Cavs net points by player

120

llimllib @billmill.org · 02/05/2026

versus the Sixers - everybody is more up and down except PG, who does what he does every night

000

llimllib @billmill.org · 02/05/2026

Celtics players' net points in the postseason. Incredible how Tatum has been able to come back and be clearly the best player on the team so far

A graph showing Celtics players points per game in the postseason

331

llimllib @billmill.org · 10/04/2026

mdriver does pretty okay on your file I think

a screenshot of mdriver outputting a similar markdown file

110

llimllib @billmill.org · 28/03/2026

Picked up Veedon Fleece by Van Morrison and enjoying some peaceful listening, thinking that not all things are bad

A record sleeve on top of a record player

030

llimllib @billmill.org · 26/03/2026

Last night I made a tool for creating player comparison graphics based on ESPN Analytics data. Here's Julius Randle vs Karl Anthony Towns; you can also expand it out to look at different play types Compare any two players here: billmill.org/nba/compare#...

A graphical comparison of KAT and Julius Randle by some different components

040

llimllib @billmill.org · 26/03/2026

Fixed up the "3-point defense" and "2-point defense" columns, which were incorrect data

A statistical comparison between Donovan Clingan and Rudy Gobert

000

llimllib @billmill.org · 26/03/2026

Added a little headshot overlay effect, that's probably good enough to ship it

comparison of Jaylen Brown and Shai Gilgeous-Alexander by their net points components

100

llimllib @billmill.org · 26/03/2026

Working on a widget for comparing two players by their @deanolytics.bsky.social net points components

A comparison of Victor Wembenyama and Nikola Jokic by ESPN analytics net points components

130

llimllib @billmill.org · 25/03/2026

I was watching that game, but was young... this is the one for me though

120

llimllib @billmill.org · 25/03/2026

google sheets chooses scientific notation

A google sheets bar chart of arbitrary numbers from a little more than 1 million to a little less than 22 million

000

llimllib @billmill.org · 25/03/2026

matplotlib denominates it in tens of millions, and gives decimal numbers on the y axis

A matplotlib bar chart of arbitrary numbers from a little more than 1 million to a little less than 22 million

100

llimllib @billmill.org · 25/03/2026

Plotly denominates it in millions with the "m" suffix

A plotly bar chart of arbitrary numbers from a little more than 1 million to a little less than 22 million

100

llimllib @billmill.org · 25/03/2026

Observable plot uses the numbers without modification, but does not provide enough padding to make them actually visible

An observable plot bar chart of arbitrary numbers from a little more than 1 million to a little less than 22 million

100

llimllib @billmill.org · 25/03/2026

so I picked 10 numbers between 1 million and 22 million (not testing across orders of magnitude, I know, just large number handling here) and created a default graph with each. ggplot2 uses scientific notation

A ggplot2 bar chart of arbitrary numbers from a little more than 1 million to a little less than 22 million

110

llimllib @billmill.org · 25/03/2026

Fixed the x axis on this graph! Michigan State is the best remaining team at defensive rebounding, not the worst; Alabama and Arkansas lag behind.

000

llimllib @billmill.org · 24/03/2026

Free Throw Rate Texas, Arizona, and St. John's rely on getting to the line UConn and Houston rarely get to the line and give up a lot of freebies Nebraska and Purdue are the ethical hoopers, neither drawing nor committing many fouls

000

llimllib @billmill.org · 24/03/2026

Turnover %age Houston, Arkansas, Alabama and Illinois do the best at protecting the ball Iowa St, Houston and Iowa force the most turnovers

101

llimllib @billmill.org · 24/03/2026

Rebound %age is my favorite one, and reveals different strategies. Tennessee crashes the offensive glass best, Alabama and Arkansas dominate the defensive glass. Nebraska doesn't offensive rebound at all

310

llimllib @billmill.org · 24/03/2026

Sweet 16 teams four factors graphs; first up, eFG: Michigan has the second best eFG defense and the best eFG. Purdue scores a lot but gives up good shots, and St. John's/Tenn/Houston have great defense but trouble scoring efficiently

110

llimllib @billmill.org · 20/03/2026

its cobol: gist.github.com/llimllib/4b2...

The COBOL program (a.cobol) is a faithful translation of a.py. Here's a summary of the translation approach:

1. Input parsing — Reads from stdin line by line, splits on space, parses comma-separated numbers into arrays
2. Memoized recursive count() — Implemented as an iterative state machine with an explicit stack (since COBOL
doesn't support real recursion). Three states:
- State 0 (entry): Check cache, handle base case (no machines left), or initialize the outer loop
- State 1 (outer loop): Run one iteration of the matching loop, push state to stack when recursing
- State 2 (resume): Add the recursive result to the accumulator and continue the loop
3. Memoization cache — Open-addressing hash table with 262,144 slots using djb2 hashing and linear probing
4. Unfold — Replicates the condition string 5× with ? separators and the machine list 5×

000

llimllib @billmill.org · 20/03/2026

it took it a minute or so to come up with a program that works on small inputs, but fails on larger ones, because COBOL doesn't really do recursion

Segfault - likely a stack overflow from the recursive PERFORM calls or an out-of-bounds array access. The
simulated recursion via PERFORM is problematic because COBOL's PERFORM doesn't truly return to the caller in the
way we need for this recursive pattern. Let me rethink the approach entirely.

The fundamental issue is that COBOL doesn't support real recursion well. The PERFORM-based simulation where
COUNT-OUTER-LOOP calls COUNT-RECURSE which calls COUNT-ENTRY which calls COUNT-OUTER-LOOP creates a deep PERFORM
nesting that overflows. I need to implement this as a true iterative state machine with an explicit loop and
goto-like control flow.

200

llimllib @billmill.org · 20/03/2026

They are actually doing pretty OK with languages with limited corpuses as long as you give them appropriate documentation / examples to use. They're reasonably good at cobol. I picked a hard advent of code day, and gave it my python solution and asked it to write cobol... +

A screenshot of a terminal showing an LLM offering to translate python into cobol

130

llimllib @billmill.org · 15/03/2026

Every year I get more excited about this than everyone else does, I just love it and I don't exactly know why

000

llimllib @billmill.org · 15/03/2026

a photo of a line from Elif Batuman I liked

000

llimllib @billmill.org · 12/03/2026

Wrote an all-new NCAA bracket randomizer this year, for once I'm ready before Saturday night llimllib.github.io/ncaa-bracket...

130

llimllib @billmill.org · 11/03/2026

A plot of every 35+ point game this season

a swarm plot of every 35+ point game this season. Dot size is scaled to the number of points, making Bam's 83 point game look ridiculous

071