Kyle Thayer
LightNews LightNews
banner
kylethayer.com
Kyle Thayer
@kylethayer.com
330 followers 500 following 260 posts
Assistant teaching professor at UW iSchool. Interested in Programming, Culture, and Education. PhD from UW CSE. Co-author of Social Media, Ethics, and Automation (bit.ly/smeabook). Opinions my own. He/him.

kylethayer.com
Posts Replies Media Videos
kylethayer.com
I'm trying use clustering to separate out dialects of Spanish with our color naming data. I don't think I'm there yet, but it's fun to try.

idl.uw.edu/color-naming...

I could use more color data in any language! ~15 minute color perception survey here: studies2.labinthewild.org/color-percep...
Graph of color nodes connected with lines. Nodes are different colors, and each node has a Spanish color name. The big colors in the middle are azul, amarillo, naranja, verde, etc. Cian, celeste and turquesa are all far apart, meaning they weren't used much by the same people (in our limited sample).
January 12, 2026 at 9:16 PM
kylethayer.com
Optimal square packing is also cursed (this is 17)

www.combinatorics.org/files/Survey...
17 squares packed in a larger square. But the ones in the middle are at such inconsistent angles and overlaps that it is unsettling
January 11, 2026 at 3:05 AM
kylethayer.com
I've made a new diagram of the Oklab/Oklch color space, which I've put on Wikipedia.

Oklab color space is intended to be perceptually uniform, meaning that it fits the weird shape of actual normal human perception. Oklch is the same space but measured in circles.

en.wikipedia.org/wiki/Oklab_c...
Diagram showing cube bins of the Oklab colors, Oklch bins displayed in Oklab coordinates (so they look show up like arcs), and Oklch bins displayed in Oklch space, where they show up in rainbow order.
December 21, 2025 at 10:59 PM
kylethayer.com
I should add: we already found these naming differences in a 2019 paper by Younghoon Kim, me, Gabriella Silva Gorsky, and @jheer.org

idl.uw.edu/papers/multi...

The new visualizations have: more data (and more languages), use Oklab/Oklch instead of CIELAB, and have new binning strategies and views
Many LAB bins in of colors shown divided into colors in English and Korean. English has 10 clusters corresponding to basic English color terms (black, brown, gray, purple, red, green, blue, orange, pink, and yellow. Korean has 16 clusters: 검정색 (black), 갈색 (brown), 회색 (gray), 빨강색 (red), 남색 (navy blue), 파랑색 (cobalt blue), 하늘색 (light blue), 자주색 (dark magenta), 보라색 (purple), 연보라색 (lavendar), 청록색 (blue green), 초록색 (kelly green), 연두색 (spring green), 주황색 (orange), 분홍색 (pink), and 노랑색 (yellow)
November 26, 2025 at 4:04 PM
kylethayer.com
The bins in all these diagrams are equal perceptual size (equal volume in Oklab space). See my diagram of the relationship between Oklab and Oklch.

For each language, the tiles are sized based on the percentage agreement on the most common color word for that tile.

en.wikipedia.org/wiki/Oklab_c...
Color tiles. First Oklab cubes shown as multiple color ranges at each brightness level. Next is Oklch bin shown in Oklab color space, but the bins are in rings around the center (c is distance from center which is saturation, h is angle, which is hue). Finally there are the same Oklch bins shown in Oklch color space where each set covers the hue range of rainbow colors + purple on the x axis and lightness on the y axis. Each set is more saturated hue.
November 24, 2025 at 5:13 AM
kylethayer.com
Weird, uploading images worked from Chrome but not Firefox. Let's demo this to see. This is from Firefox. Let's see if it works now.

@pfrazee.com (is there a better bug reporting line?)
Should be a bunch of color tiles showing color naming in English and Korean
November 24, 2025 at 5:05 AM
kylethayer.com
I'm excited to have updated our color naming in different languages project to use the new Oklab/Oklch color space.

The new views highlight the differences in how English and Korean divide blues, greens and purples.

idl.uw.edu/color-naming...

@jheer.org @uwcse.bsky.social @ischool.uw.edu
A hue band (in Oklch, equal in c) of colors shown divided into colors in English and Korean. Where English has one "green," Korean splits into "연두색" and "초록색." Where English has one "blue," Korean splits into "하늘색" and "파랑색." Where English has one "purple," Korean splits into "보라색" and "연보라색" and "자주색." Many hue bands (in Oklch, each is equal in c) of colors shown divided into colors in English and Korean. Where English has one "green," Korean splits into "연두색" and "초록색." Where English has one "blue," Korean splits into "하늘색" and "파랑색." Where English has one "purple," Korean splits into "보라색" and "연보라색" and "자주색." Many rings of color bins (in Oklab, equal in l) shown divided into colors in English and Korean. Where English has one "green," Korean splits into "연두색" and "초록색." Where English has one "blue," Korean splits into "하늘색" and "파랑색." Where English has one "purple," Korean splits into "보라색" and "연보라색" and "자주색."
November 24, 2025 at 5:03 AM
kylethayer.com
A few weeks ago I saw a flock of crows rip one of the UW fields of grass to shreds. So UW put in seeds and a "keep out" barrier around it, but apparently the crows are back and they don't respect the sign 🤣

(There were only 5 crows this morning, while the other day I saw dozens)
Damaged field of grass with rope barrier and sign "Please... KEEP OUT Landscape Restoration In Progess." In the middle of the damaged field are two crows ripping out sections of grass.
November 6, 2025 at 5:44 PM
kylethayer.com
I'm enjoying the new Fiona and Cake season 2 episodes!
Beautifully crudely drawing of a cartoon forest with two characters walking in it
November 1, 2025 at 6:49 PM
kylethayer.com
A student told me about the new Oklab color space, which fixes a problems with previous perceptual color spaces (like LAB).

Unfortunately, when I tried Oklab instead of LAB to make color bins for a datavis, the lightness dimension is much longer, which makes it hard to show a lot of bins.
Diagram of "All RGB Colors Binned Into Cubes." Shows bins in Oklab space and in CIELAB space. Each one is divided into 11 lightness level (Oklab: 0, 0.1, ..., 1; LAB: 0, 10, ..., 100). At each level is a set of squares along the a and b axes. In Oklab each lightness level is about 5 by 5 bins, but in LAB it's about 15 by 15 bins).
October 31, 2025 at 9:46 PM
kylethayer.com
Yes, substack has a share option to share to Bluesky which includes that suggested post text.

(I've done share buttons with suggested post text before for research surveys where we wanted to encourage people to post the survey on social media)
substack article with a share button with options to share to Facebook, LinkedIn, Bluesky, etc. Auto-filled in draft of a post on bluesky that appeared when I pressed the share button on substack
October 29, 2025 at 9:03 AM
kylethayer.com
I just noticed we see the same pattern in the misspelled "chartruse" and "chartruese!"

I'll have to update our data cleaning rules later to group those misspellings with "chartreuse."

idl.uw.edu/color-naming...
Color ranges for misspelled "chartruse" and "chartruese": mostly a dark greenish color but sometimes a reddish purple color
October 7, 2025 at 4:21 PM
kylethayer.com
The English color term people most disagreed on was "puce." Our closest matching label was "idk."

idl.uw.edu/color-naming...
Color range for "puce" which ranges from green to purple to gray to brown.
October 7, 2025 at 4:03 PM
kylethayer.com
When we asked people to name color tiles, this is the range of colors people called "charteuse," which includes the color you said. So there are certainly some people who agree with you.

idl.uw.edu/color-naming...
English chartreuse color. A set of tiles show the range of colors given the name "chartreuse" which is mostly a pale or dark green, but also a dark red.
October 7, 2025 at 3:54 PM
kylethayer.com
If most people agree on the wrong spelling fuchsia wrong, does that make it the new right spelling?

When we asked people to name random colors, the dictionary spelling of fuchsia came in second place to "fuschia."

We probably missed even more variant spellings in our data

bsky.app/profile/kyle...
Graph of Variations of Spelling of Fuchsia. They are, in order, "fuschia" 744 times, "fuchsia" 591 times, "fuscia" 258 times, "fushia" 181 times, "fusia" 74 times, "fucsia" 29 times, "fuchia" 29 times, "fusha" 15 times, and "fuchsia?" once.
September 28, 2025 at 3:28 PM
kylethayer.com
Since the preview doesn't show it, here's the diagram I made for Wikipedia:
September 21, 2025 at 10:54 PM
kylethayer.com
"They say you gotta spend money to make money, well I dunno where we went wrong! We spent ALL of our money!"
September 8, 2025 at 5:32 AM
kylethayer.com
For context, Rebecca Sugar fought for the queer wedding episode and had to accept a compromise to rush the end of the Steven Universe main series.

The follow up movie and Steve Universe Future were titled differently so they could be placed in overseas markets as shows-with-no-queer-weddings.
Ruby and Sapphire's wedding in Steven Universe (both Ruby and Sapphire are female presenting rock-aliens, Ruby is red with short hair and a dress, and Sapphire is blue with long hair and a suit)
September 1, 2025 at 3:29 PM
kylethayer.com
Me and @susannotess.bsky.social make sure to cover that "All data is a simplification of reality" early in our textbook as students are learning about programming data types.

Like this photo showing 2 apples + 2 apples = 4 apples

social-media-ethics-automation.github.io/book/bsky/ch...
photo of the two regular-sized apples from before, but now next to two tiny apples
August 1, 2025 at 10:12 PM
kylethayer.com
I spent some time this summer reading about how speakers work ("driven damped harmonic oscillators," "minimum-phase," etc.) for this (hopefully) accurate diagram.

Now it's on Wikipedia en.wikipedia.org/wiki/Loudspe... and my textbook social-media-ethics-automation.github.io/book/bsky/ch...
July 31, 2025 at 5:52 PM
kylethayer.com
I'm reading "Rhetorical Code Studies" by @brockoleur.com and thinking about isomorphic code (code that is the same algorithmically, but is named differently and has different meanings).

For example, these are all structurally the same, but have different meanings and even political implications:
Screenshot of code: function multiply(a, b){ return a * b; } function getCircumference(radius, angleInRadians){ return radius * angleInRadians; } function totalWeaponsCost(weaponCost, numWeapons){ return weaponCost * numWeapons; } function totalFlowersCost(flowersCost, numFlowers){ return flowersCost * numFlowers; }
July 22, 2025 at 10:45 PM
kylethayer.com
In recognition of what PBS and NPR funding have done, here's a clip of Fred Rogers testifying before congress in 1969 to get federal funding.

Full video here:
misterrogers.org/videos/pasto...
July 18, 2025 at 12:46 AM
kylethayer.com
🤣
L (Liability): Model Collapse. Models degrade if frained on synthnetic data. M (Model Complase): Model Collapse. Inimors riset. frent people and benefits due atsax.
July 14, 2025 at 10:19 PM
kylethayer.com
I doubt it is an intention to reduce copyrighted responses. Generative AIs are just extremely prone to producing nonsense. Like here is ChatGPT labeling an upside down map:
A labeled map of the world rotated by 180 degrees, but Australia is “North America,” South Asia is “Mexico,” China is “Africa,” South America is “Australia,” the Atlantic Ocean is “Midle East,” The USA is “Australia,” etc.
June 13, 2025 at 4:22 PM
kylethayer.com
Bunch of State Troopers rolling through University of Washington campus right now. Anyone know what that's about?
A dozen or two State Troopers including a car going through the pedestrian area near the library / fountain
June 12, 2025 at 8:28 PM