Inspiration: Curiosity initially around what the most common crossword layout is. Kinda morphed into wondering about distributions of empty spaces and letters.

Description: I downloaded some crossword data from Saul Pwanson. I focused on the NYT crossword which has had two variations for its history, the 15x15 daily and the 21x21 Sunday version. For each version, I tallied up counts of characters per cell.

What's visualized above are frequencies that answer the question “When character X shows up, where is it most likely to show up on the board?" The shading is scaled to the max per-cell-frequency of that particular character. So it's really highlighting distribution, not overall frequency.

For example, the character A is far more common than G, but G appears darker because it's very evenly distributed whereas A has a couple of extreme outliers in the top left. It's fun clicking through as some patterns emerge.

Quasi-Groupings

Anywhere works

A    G    K    L    O    R    T    W

Great starters

B    C    F    M    P

Great enders

D    E    S    Y

Anywhere but the start

N

Anywhere but the end

H    I    U

Just the middle

V

Rarities

J    Q    X    Z

Sources

NYT Crossword Data (2024)

But I think my favorite part has been the repeated opportunity to wander. I love to be aimless, love the feeling of discovery.

After our latest move to another city, I'd been trying to find fun things to do with my 1 and 2 year old kids: parks, playgrounds, libraries.

I find it harder to be freeform with such little kids - having a plan puts me (and I think them?) more at ease. So naturally, I turned to the Internet and found some great things. Walkable spots. Close-by drives.

After a month of being here, though, I finally took some time for myself to just go on a random walk. No particular purpose. I ended up passing through the little city center, kept going into the town gardens that then turned to walking paths among rows of beautiful trees.

I took a left on a whim. It was foggy and through some pines, I saw a low fence, then a slide and swings and realized there was this magical little forest playground just sitting there. I was delighted. Couldn't wait to walk the 15 minutes back home, tell my kids and bring them over!

Later, I wondered how I'd missed this delightful spot, and it was simple. Just wasn't on Google Maps for whatever reason.

And that's the beauty of wandering. I could've easily settled into my planned routine, the places I'd already visited, and that would have sufficed.

I also could have done a more in-depth online search or looked at Google Maps satellite view, written a program to detect unknown playgrounds etc etc.

But the sheer serendipity of opening up to life can be so surprising and fulfilling. It brings me immense joy and peace to stumble upon some little marvel.

Sometimes, finding a clearing is as literal as happening upon a space set aside in a dense forest. Or finding the vast horizon at the top of a tree. Sometimes it's taking a few deep breaths, emptying your mind for a millisecond of calm.

Inspiration: Desire to learn some more about LLM outputs and the 1:1 temperature mapping to the warming globe was a salient analogy for things melting down

Description: I repeatedly asked an LLM to complete the sentence “Climate change is _____ at +[0 - 2.5]°C". And each time, I increased the LLM's response temperature (increases randomness) from 0 to 2.5.

This maps to the global warming temperatures above pre-industrial levels from 0°C to 2.5°C, a projected temperature we'll likely hit by the end of the century. As of 2024, we were around 1.55°C above those levels.

Method: I wanted to play with LLM outputs so I ended up using ollama with llama3 locally to have unfettered access to parameters like temperature, top_k and top_p. Temperature alone didn't drastically vary outputs for such a simple prompt. So I also increased top_k from 40 to 10,000 to capture unusual words and increased top_p from 0.9 to .99 to make sure the unusual responses had a chance of being returned.

I found the simple 1-word fill-in-the-blank outputs easy to compare as the model repsonses melt down under increasing temperatures. I'm interested in the odd long-tail outputs of these models. The fabricated portmanteau of “grimmanent" (grim + immanent) was fascinating for instance.

Also a reminder that we, both end users and model builders, barely scratch the surface of possible outputs. There is an art to what LLMs can present us and no guarantee it's responding optimally per prompt.

Sources

Ollama docs

- Mary Oliver, wonder of a poet

Makes me think about all the poems I've written.

Maybe 1 out of every 100 or 200 scrawls, I find it. The right word, the delightful phrase, something utterly true.

I don't mind being a broken watch - it's all about volume. Try. Create. Fill the minutes with life, discovery.

Makes me dream about all the poems I'll write.

- T.S. Eliot

Since all I can focus on are the game stats periodically ticking, I started noticing patterns. Like how often the Pacers and Thunder, as entire teams, would hover above or around 50% field goal, 40% 3-pt and 90% free-throw percentages. Each team even finished a game hitting that 50-40-90 mark. Super efficient stuff.

We like to keep track of the 50-40-90 club for season averages of (mostly) legendary individual players: Bird, Reggie, Nash, Dirk, KD, Steph et. al. So I figured I'd take a deep dive into teamwide 50-40-90's and write up what I find1.

Disclaimer: I'm aware there are many better ways to describe a team's shooting performance nowadays, but I enjoy the heuristic. I added true shooting percentages to the tables to provide some extra context.

The Playoffs

Let's start off by looking into the playoffs. The 3-point line has been around since 1979-80 in the NBA, and since then we've seen 64 instances of a teamwide playoff 50-40-90.

Teams hitting those marks went 59-5 (.922) - admittedly more losses than I anticipated, but still very, very good. Teams hitting the marks in the Finals went 3-2 (.600).

Turns out the Thunder and Pacers are the only teams to hit 50-40-90 in the playoffs over the past 2 years. Should lead to some impressive shot-making in this year's Finals.

Interestingly, the Phoenix Suns lead all franchises with 10 playoff occurrences of the teamwide 50-40-90. All that offensive firepower leading to zero titles is perhaps an indictment against chasing these numbers...

The Regular Season

In the regular season, the teamwide 50-40-90 has happened 1041 times - around 1% of games in the time period - leading to a record of 909-132 (.873).

I don't know why I thought it would be even more of a guaranteed win hitting those numbers - guess I was anchored on a somewhat undeserved awe of the thresholds. That win percentage would translate to a 71 or 72 win season though, which is pretty dominating.

Interestingly, the Indiana Pacers lead the league with 61 occurrences. Sadly, these offensive displays have also led them to exactly zero titles. Although maybe, just maybe, they can fix that this year.

An Artifact

With so many 50-40-90 occurrences, it turned out that 7 times, both teams hit those marks in the same game. They've mostly happened in the past decade, speaking to the increase in efficiency and league-wide shooting talent.

In this mini dataset, I still found it odd that the 2018 Warriors could have lost any game in which they went 50-40-90. I clicked into it and looked at the box score. Not only did Steph and Klay not play, but KD also got ejected less than 20 minutes into the game, after shooting poorly.

The weird part then became the fact that they still hit the 50-40-90. And it happened mostly because of a sharpshooting Quinn Cook who went 12/15. Digging further, it turns out Quinn is the sole and founding member of the G-League 50-40-90 club. So in a full-circle way, it made sense after all.

A More Exclusive Club

I originally thought the teamwide 50-40-90 was going to be rare. With 1000+ occurrences (~25 per season), I felt sort of underwhelmed. So I've upped the stakes here and filtered for a more exclusive 60-60-90 club.

It's an order of magnitude more difficult to accomplish with just 14 instances so far, and none in the playoffs. As an avid Mavs fan, my favorite part of the list is that a Dirk + Kidd + the JET + Barea team qualified.

My second favorite part is that somehow the pre-LeBron Cavs joined this club, but were the only ones to lose the game. In 2016, the Chosen One really did break a nasty franchise curse.

The Platonic Ideal

A version of the perfect teamwide 50-40-90 game would be that every single player who attempts shots also hits those thresholds.

Crunching the numbers, I didn't find any games that quite met this player-by-player elite shooting display. I found a few games that came fairly close, even exceeding the marks in some categories.

• 2000 New York Knicks: 40-50-100
• 1990 Charlotte Hornets: 40-40-100
• 2025 Miami Heat: 40-30-80

Maybe in a decade, I'll check back on this and see whether the elusive club has any members.

References

1. Data from Kaggle NBA Dataset - Box Scores & Stats (1947 - now). I used gocsv + Python for data processing

In the past couple years, led by Anthropic, a handful of LLM-related websites have co-opted the design style. Their branding in general aims to give the machine an implied humanity. Sometimes I wonder if that's how they've tricked me into using Claude more than other models...

Anyhow I've collected some examples of the beige website here and I'll keep adding them as I find them. Admittedly, a couple might lean more gray when I see them in context here.

Examples

Inspiration: The incredible trees that I've had the luck to climb in my life

Description: Wanted a place to collect some of my favorite trees. Living in Northern California for about a decade, I was able to free-climb a bunch of redwoods, pines, oaks, and fruit trees. I also added in some favorites from my time in Puerto Rico (flamboyan) and Texas (live oak). There's nothing quite like the feeling of acension, leaning to hang with the wind and the leaves and the sun.

Legend: The graphic below is a phylogenetic tree of trees. It's a stylized legend to help understand how closely (or not) these trees are related to each other genetically. The oldest and least related to the others is the Redwood tree (240 Million years old), while the two most closely related are the Lemon and Orange trees, two of the youngest trees (~20 Million years old).

Sources

GenBank - phylogenetics

PhySortR - plot

Inspiration: (1) The Hungarian-monk-founded Cistercian school I went to in Texas and (2) the code and cool project page on Cistercian numerals by Ákos Nikházy

Description: It turns out the Cistercians (an order of Catholic monks) created a numeric system that could represent the numbers 1 - 9999. The ciphers have nice symmetry and, with some practice, become quick to read. I annotated them up top to show the locations of the ones, tens, hundreds and thousands. A quick clickthrough has a fun sort of stick-figure flipbook aesthetic. Probably what they were going for with these runes...

Legend

Sources

Ákos Nikházy - canvas code

Inspiration: A series of beautiful maps created by Chaney Swiney, composed with excellent colors, typography and framing

Description: Felt like this was such a cool type of map that I wanted to try my hand at it. Focused on some of the islands that mean the most to me. I think the original maps were made in Natural Scene Designer by starting with a Distance Mask effect. I wanted to order the product, but it seems to be CD/DVD based - a wild flex in the year 2025. So I pivoted to what's good enough and free - got an approximately similar effect using the QGIS DEMto3D plugin combined with Blender's Mist Pass - may also be worth investigating distance to camera + textures?

Sources

Chaney Swiney - peakscapes

Open Topography - DEM

data.gouv.fr - DEM

Inspiration: The fascinating, sometimes disorienting, image processing by the Japanese artist Kensuke Koike (see image below)

Description: So much of the statistical world involves building a model: reducing reality to a more manageable number of dimensions. Even the billions of parameters in the largest LLMs are a simplification of the sheer complexity of reality.

I think the rearrangements of Kensuke Koike provide beautiful intuition into the approximation of what's real. The work creates something wonderfully new, yet remains reminiscent, yearning, incomplete. A jumbled hint. I tried to recreate this effect in code and applied it to some of my favorite paintings and artworks.

Sources

Kensuke Koike - visual artist

Public Work - public domain artwork

Inspiration: (1) The beautiful, mazelike, single-line drawings by the Polish architect and engineer Wacław Szpakowski (2) the serenity of Japanese zen gardens and (3) these kinetic art sand tables by Sisyphus

Description: Wanted to make something slow and peaceful, developing on the page. Each animated drawing is a single, uninterrupted line. And all the drawings are vectorized versions of Wacław Szpakowski's original works. I kinda gave up (for now) on a programmatic way to trace his lines that maintained the sharp corners and straight lines. I tried some out-of-the-box Python skeletonization libraries, medial axis transforms, etc but nothing quite got it. So what remains are squiggly vectors derived from Inkscape's “Bitmap Trace - Centerline" feature.

Sources

SVG Artista - very cool!

Devina - SVG minifier

Inspiration: My little 1-year-old walking around, discovering brand new sounds and words and formulations - really cool to transcribe some of her original words : )

Description: There are thousands of known languages out there which is incredible. Watching my kids learn to speak, it's fun to imagine the literally billions of fleeting dialects that children all around the world express as they discover language. So for now, I'm trying to capture these novel words / sounds and just encourage my kids to be playful and inventive with words.

<p><b>Languages</b></p> 

French → Fre.

Spanish → Spa.

Tahitian → Tah.

English → Eng.

Unknown → Unk.

Sources

ISO 639-3 - language codes

Inspiration: The Oxford Dictionary of English Etymology edited by C.T. Onions- such an incredible 1800's name

Description: Word origins fascinate me, and I wanted to discover the etymologies of all the words I write here on my website. Some words are simple - the word area in English comes from the word area in Latin. Some words are thornier - the word unbelievably is forged from (1) un- (2) believe (3) able (4) -ly, which each have their own multiplying, jagged branches into the past. Cool to explore, time-consuming to visualize. So for now, I stuck with visualizing the simple words - the ones on my website with straight line, direct etymologies all the way back to their origins.

Languages

Latin

Old French Middle French French Spanish Italian Portuguese Old Provençal

Egyptian Hebrew Arabic

Ancient Greek

English Middle English Old English Scots

Dutch German Old Dutch Middle Dutch Old High German Middle Low German Yiddish

Old Norse Norwegian Proto-Germanic Anglo-Norman

Sanskrit Hindi Gujarati Kannada Persian Pahlavi

Russian

Mandarin Chinese Japanese

Malay Hawaiian Tahitian Taino

Romany Angloromani

Engenni

Unknown

Sources

Etymological Wordnet- dataset

ety-python- library to access data

Inspiration: The limit of a regular n-gon as n → ∞  is a circle.

<p><b>Description</b>: In the physical world, a perfect circle does not exist and cannot be drawn. In the world of pixels and perception, however, there is some point at which our eyes no longer discern the <i>n</i>-gon. I think, for this graphic, I see a circle around the <i>85</i>-gon and up. Very dependent on size, canvas resolution, widths, color etc. Zooming in, the illusion breaks down even at the hectogon (<i>100</i>-gon). Was also fun familiarizing myself with some of these unusual names for regular polygons : )</p>
<p style="margin-bottom: 10px;"><b>Sources</b></p>
<p class="indent"><a href="https://www.kirupa.com/canvas/easily_draw_any_polygon.htm" target="_blank">Kirupa</a>- code for drawing <i>n</i>-gons</p> 
<p class="indent"><a href="https://en.wikipedia.org/wiki/List_of_polygons" target="_blank">Wikipedia</a>- regular polygon names</p>

Inspiration: Exhibit at Musée de Tahiti et des îles

Description: Just wanted to have fun animating letters, trying to come up with something interesting, slow, evolving. Maybe evoking synaesthesia? The colors of the letters (graphemes) correspond to the categories of ARPAbet consonants / vowels (phonemes) in the legend below. The mapping of graphemes to phonemes comes from some of my previous (non-academic, non-official) explorations.

Legend

plosives: BDGKPQT

fricatives: DHFVSSHTHZZH

nasal: EMENMNNG

affricates: CHJH

liquids: ELLR

glides: WWHY

whisper: HH

front: AEEHEYIHIY

mid: AHER

dipthongs: AWAYOY

back: AAAOOWUHUW

There are dim pockets in the woods, quiet places. Tree bark coursing with ants, crumbling branches, dark and humid. There's an allure to their stillness, unmistakable smell. Organic residue and soft ground.

Forests tolerate them for years, decades - sites of latent resurrection. In the meantime they are mapped by those within.

Eerily calm as they writhe, toil. Decay is a series of cliffs, sudden breaks into canyons.

Undergrowth

I want to trace the line, electric. With another sense, myopic. There is a net, a flow, resilience, springy with the strength of life. Wily, clever little lines.

Can you build an essay this way, a poem? Should I dig up the supports? Less a maze, perhaps a village-turned-city-turned-upside-down. I don't know if time is cyclic, but matter has many methods of rebirth.

Scale visits this dwelling, then turns away. Moonless, sunless, vestiges appear and fade. A striving, certain.

Canopy

Illusion of breath, spaces between spaces / species. Thin workhorse, leafvein transporting pristine light. Lithe bend of top branch, far-reaching spirals.

It seems we have conquered death into dainty blooms. Suffocating green film. Maximize capture, minimize competition.

It is, again, mathematics, or some theory of it.

We try and try and try to devour content as quickly as servers erase latency. But when we act like computers, fast and ceaseless, it is exhausting, demoralizing and comes at a cost.

Our online brains cannot truly handle the ever-present click, swipe, dopamine, turn to the next novelty. We've drifted into false time and we're floundering. I don't think we were built for this.

Instant gratification

Many websites, apps, streaming services etc. aim to offer you XYZ in “real time”, which really just means “instantaneously.”

Which really just means “we-desperately-want-to-give-you-what-you-want-without-a-single-moment-to-think-because-if-you-think-you-might-leave–and-your-attention-is-our-money.”

Companies don't always succeed, but their goal is clear. Any time spent buffering, waiting, etc. could be a death knell for their product / service.

The user could take their tech-induced craving for instant gratification to some other product, or worse, a competitor.

The online reality this has created is a very frenetic one.

Human time

But real time, our shared, collectively perceived time, isn't like this. Real time flows in different ways. Sometimes it actually is immediate sensory feedback.

Yet often, it draws patience out of us. Vines bloom into flowers, smiles press into wrinkles. Florescence, senescence. This is also our real time of decades and months.

So don't be afraid to step away from this screen, give up the race with computation and explore some other direction. Find comfort in your humanity and find relief in the slowness of things.

References

1. My mom, who wisely preaches balance and delayed gratification - useful things in a hectic world

2. T.S. Eliot's Four Quartets

You and I, will, on some level, always be guessing. We don't have all the variables, we never will, and even if we did, the future will be irreconcilable from the present, and especially from the past.

Peering forward

Humanity has a long history of searching for an oracle. We are a collectively curious, but anxious species.

Religiously, we've looked for telltale signs, prophesies or omens. Scientifically, from the dizzying matrix of existence, we've strung out threads of knowledge. And some of these are even powerfully predictive (e.g. movement of celestial bodies).

Despite this progress, the vast future across all domains remains, in the truest sense, unknowable. Why? There is irreducible uncertainty in systems that can only be modeled probabilistically.

Physical systems

In the physical world we have to deal with a hazy cloud of possible outcomes all the time. Weather systems are a prominent example.

Despite knowing how these systems work on a physical level, their behavior is nonlinear and is very, very sensitive to initial conditions. We can work on the reducible uncertainty portion of this problem, i.e. add more sensors, collect more data points in the atmosphere, the oceans, etc.

This helps us better understand the initial conditions, and helps predict what will happen in the near future with a little more confidence. However, the uncertainty compounds exponentially the farther you move away from those initial conditions, quickly dissolving confidence in our predictions.

Computers

There are many systems that don't behave nearly as chaotically as weather, yet remain subject to irreducible uncertainty. The computer is one of those systems. It is an unbelievably reliable invention.

We can almost always predict that the computer will do what we tell it to do. Yet, on its most essential level, at the bitwise operator, a computer can be affected by chaotic natural phenomena.

The typical way a computer stores bits is with a transistor gate, marking the presence or absence of voltage as a 1 or 0, respectively. If rogue electrons flow through this gate, they could unexpectedly flip the bit from a 0 to a 1 or vice versa. Too much heat or faulty hardware can cause such an error, known as a bit-flip.

But so can the magnetic field created as cosmic rays hit the transistor. Because outer space is awash in cosmic radiation (and other risk factors), spaceships minimize risk with redundancy in their physical computer systems, sometimes employing 5 computers.

If there's a discrepancy in the expected output at any given time, the computers tally what their outputs are showing, and proceed with majority agreement.

While the frequency of these cosmic rays can be modeled probabilistically, the observer remains fogged in a gray area of uncertainty. We'll never know exactly when one of these randomly occuring events will actually occur.

So even for our most robust input/output machine, operating on a millisecond-into-the-future time scale, we cannot be 100% certain it will provide us with the future outcome we predict / expect. Wild.

Moving forward

Sometimes, it can feel like we're gaining immense amounts of knowledge, so much so that the future is in our grasp. While we've absolutely made progress peering a little ahead, it's important to recognize that any prediction is wrapped in uncertainty.

It can be humbling to realize this, but ultimately it's also the source of our freedom, our ability to stray from predetermined lines and squiggle our own paths into the future.

References

1. Stephen Wolfram's interesting take on the phenomenon of free will and how it relates to computational irreducibility

2. Vinod Wadhawan's blog discussing natural phenomena and Laplace's demon

Our present century brought the massive expansion of the internet, for better or worse. In order to grapple with its prominence, people have, unsurprisingly, looked to the trees for understanding.

The dark forest theory

Perhaps in the early years, the online world grew organically from individual gardens and seedlings into a forest with enough light, variety and space to enjoy a walk in the woods.

People tended to their plants and trees with care and sincerity. They placed personal interests and hobbies out there and people began to find unexpected community. Though there was always sketchy activity, these dangerous trails were easier to avoid.

Soon after the people, however, came the money. It was clever. It was voracious. And it has devoured us. Nowadays, to move through the open internet is to be monetized or scammed.

There's a rot in this aging forest, parasitic growth choking the trees, threats lurking. The trails are overgrown, full of sticker weeds, thieves, traps and poisonous vines. We've entered the darkest age of Mirkwood.

This transformation is at the core of Yancey Strickler's Dark Forest Theory of the internet.2 The invasion of ads, data collection, dark patterns, harassment, negativity-driven engagement and bad actors has become so overwhelming that we don't even want to make a sound, lest we draw public, destructive attention to ourselves.

It's no wonder we've retreated underground to more private spaces, to Venkatesh Rao's cozy web (messaging apps, Discord, Slack, email, etc.).

Extending the metaphor

Maggie Appleton builds on this theory by describing how generative algorithmic content may send the digital world into utter chaos. We now have Rogue Johnny Appleseed and his many clones planting a practical infinity of LLM-driven, derivative trees.

The forest was already vast and darkening. Now it will likely become even more immense and treacherous. If a core purpose of going online is to connect with other humans, that's about to become much more difficult.

In response, it's tempting to fully disconnect or abandon the online world. I think this will / should happen to some degree and doing so will probably benefit many people (e.g. by eliminating doomscrolling, etc.).

If you want to trade metaphorical trees for natural trees, you'll find no argument here. I cherish every visit to the forest and every climb into the canopy. I know how it heals and refreshes me.

What's at stake

If you want to stay online in any capacity, I believe cultivating a human-centric online experience is essential. There are many possible solutions that Maggie provides to help us navigate this new reality and prove our humanity online (reverse Turing test).

I especially like her recommendations to be as original as possible and to play with unique language/expression. Though I know these aren't strict safeguards, I find a lot of promise there.

It'll be tricky, but if we unknowingly lose ourselves in a thicket of LLMs, I'd find that rather tragic.

I care about the triumphs, ramblings, learnings and misadventures of humanity, not the muddled, plagiarized, probabilistic output of algorithms.

Genuine human intelligence

The ongoing transformation is partly why I've started this website and why I'm writing a little. I've been a huge beneficiary of all types of human-produced content both online and offline. I've learned countless things from people and their blogs / courses / code, but I've mostly been a passive consumer.

Maybe now, disappointed by the coming changes, I want to actively contribute to a humanist vision of the internet. LLMs and other types of algorithms / statistics can and will play an enormous part in our future.

I just don't want these tools to obscure our reality or distract us from original human thought and experience. I want to participate in a public exchange of ideas, and help maintain a network of genuine human intelligence.

Reviving the forest

People often say “you don't want to miss the forest for the trees.” In other words, don't get so caught up in the details that you lose track of the big picture. In this case, the big picture is a bit of a mess. But it can be cleaned up.

I believe this online moment requires immense focus on the trees, one-by-one, in order to salvage the forest. We can make a home again in the woods. It'll take a sort of inversion, a fresh return to the nostalgic roots of the internet.

This is the start of my contribution to that endeavor, a hope to create lively new growth. And to connect with others trying to do the same.

References

1. Forests: The Shadow of Civilization by Robert Pogue Harrison - a review of the forest as metaphor in Western thought

2. I found out about these ideas through Maggie Appleton's awesome website - she's a big advocate of creating your own online garden, and I gotta say, it's been fun starting one up