iiinova — where your digital self keeps evolving

Johan Ugander @jugander.bsky.social · 18/08/2026

See tweet. Platforms change over time, sometimes a lot. These studies are a lot of work and thus a lot of work to replicate. I'd love to see better longitudinal tracking of amplification trends. Cf. Kevin Munger's work on temporal validity. x.com/nikitabier/s...

Tweet by X product manager describing changes to the platform.

031

Johan Ugander @jugander.bsky.social · 16/04/2026

Paper is full of great detailed examples of "shifting goalposts":

011

Johan Ugander @jugander.bsky.social · 09/02/2026

It's been cold in New Haven for a while. How cold? How long? What about other cities? I used claude code to cook a webtool to explore what I am calling the "cryofront": the minimax curve of how cold it's been, for how long. Feedback welcome! (AI coding is wild y'all.) jugander.github.io/cryofront/

2114

Johan Ugander @jugander.bsky.social · 03/02/2026

Most unusual SI/methods write-up I've seen in a while: Digging in Utah scrapbooks for hair samples to document historic lead levels. Remarkable findings, note the log-scaled y-axis.

1267

Johan Ugander @jugander.bsky.social · 25/01/2026

The replies to this post surfaced lots of great 3rd party tools (thanks!), but I still think it's interesting that bsky, as a platform, has decided to not prioritize algorithmic triangle closing (or follow recommendations, generally). See this fig from Zignani et al. (2014), when FB introduced PYMK.

3262

Johan Ugander @jugander.bsky.social · 08/01/2026

Here are the stats for my own account. (Separately: used this script rewrite to test drive claude code for the first time, amazing.)

010

Johan Ugander @jugander.bsky.social · 18/12/2025

Your code repo is amazingly clean, so I was able to do straightforward code modifications and replicate your Table 6 and 7 with added mayoral cohort controls. Happy to report the law enforcement effect finding in Table 6 is largely robust to this change in model spec, 3.6% instead of 3.9%. 2/

230

Johan Ugander @jugander.bsky.social · 05/12/2025

I'm just saying one could, not that one should. Often there's a lot of things one shouldn't do… including community detection ;-)

110

Johan Ugander @jugander.bsky.social · 04/11/2025

New office trinket: I 3D-printed "Lewitt's missing cube". Quite happy with how it turned out. Many thanks to Paul for sending me the stl file! Skip to 49:55 in his video to appreciate why this is fun.

040

Johan Ugander @jugander.bsky.social · 24/10/2025

Connecticut is approach peak fall. Shots from Lake Wintergreen this morning. A glorious start to the day.

1220

Johan Ugander @jugander.bsky.social · 08/09/2025

My award for most honest paper goes to: arxiv.org/abs/1804.02767 An arxiv-only paper describing v3 of YOLO, a vision algo that already had two published papers. With huge adoption already, authors were out of sell mode and just documented it as best they could. Sec 4 takes honesty to 11. 37k cites.

041

Johan Ugander @jugander.bsky.social · 03/09/2025

Google autocomplete's behavior here reminds me of @msalganik.bsky.social's very nice discussion of "algorithmic confounding" in Bit by Bit (2.4.2): www.bitbybitbook.com/en/1st-ed/ob... and @davidlazer.bsky.social et al.'s "Parable of google flu trends" (2014) paper: www.science.org/doi/10.1126/...

080

Johan Ugander @jugander.bsky.social · 03/09/2025

Google autocomplete is asking the important questions.

160

Johan Ugander @jugander.bsky.social · 30/07/2025

Does anyone know studies of how FB/X/threads/etc downrank (or don't) external links? According to Adam Mosseri in 11/2024, threads then didn't explicitly downrank links, but notes threads prioritizes like/comments, not clicks, and says people simply don't like/comment much on posts with links.

Mosseri: We don't downrank links, but we don't place much value on p(click) and people don't like and comment on links much.

131

Johan Ugander @jugander.bsky.social · 02/07/2025

The photos from Ukraine of drone fibre optic debris remind me of Italo Calvino's Ersilia: "In Ersilia, to establish the relationships that sustain the city's life, the inhabitants stretch strings from the corners of the houses, white or black or gray or black-and-white according to whether… 1/

Fibre optical cables stretched across a field in Ukraine, left behind by tethered drones used in war.

2103

Johan Ugander @jugander.bsky.social · 01/07/2025

📣 Update: I've moved to Yale, joining the Department of Statistics & Data Science as an Associate Professor today (July 1)! The last year's sabbatical here has been great for me and my family, and I am excited for great things ahead! Come visit! Trading the wind of freedom for light and truth!

14901

Johan Ugander @jugander.bsky.social · 23/05/2025

The annual snapshots also let us see how the average distances evolved over time. We saw how the average distance came down as the graph grew in nodes and edges. This allowed a conversation with work on "densification", e.g. Leskovec et al. 2007: arxiv.org/abs/physics/... 7/

110

Johan Ugander @jugander.bsky.social · 23/05/2025

Here's the "four degrees" result, where you see that the pairwise distance distribution (for the May 2011 FB graph) is really concentrated around 4. Single countries around 3. The avg distance for the FB graph was 4.74. Sweden and Italy both average 3.9. 6/

130

Johan Ugander @jugander.bsky.social · 23/01/2025

If it is reasonable to assume a noise distribution (or testable), there's a simple rule for when mean vs. median is more efficient estimator (lower asympt var). I don't remember the derivation, but I think it may assume distribution is symmetric. From (tangential): web.stanford.edu/~jugander/pa...

100

Johan Ugander @jugander.bsky.social · 08/01/2025

Mary Beth was on the original team that developed and deployed Twitter Birdwatch = X Community Notes, and at FB before that. Insightful thread. See also the Birdwatch paper, submitted to arxiv the night before Elon took over: arxiv.org/abs/2210.15723

1173

Johan Ugander @jugander.bsky.social · 12/12/2024

LLMs playing Scattergories as a parable for questions of diversity/monoculture in genAI, from Manish Raghavan arxiv.org/abs/2412.08610 When there is competition (n>1), you want an LLM with high temperature. But not too high.

070

Johan Ugander @jugander.bsky.social · 22/11/2024

Kandiros, Pipis, Daskalakis, and Harshaw have a really Interesting new arxiv preprint on "conflict graph designs" for interference/spillovers: arxiv.org/abs/2411.10908 For GATE estimation the improvement is very significant and I'm optimistic/excited about how the ideas will impact the literature..!

1248

Johan Ugander @jugander.bsky.social · 08/11/2024

After many conversations with art practitioners and many hours in the art library stacks, this essay is the result. Things also got delightfully weird when we tried to decide how to randomize author order. I'm pretty sure this is the first paper with an author order determined by tarot! 4/4

050

Johan Ugander @jugander.bsky.social · 16/10/2024

I gathered US News "peer assessment" response rates. They appear to be at an all time low. Importantly, what explains the sudden rise in the "National Liberal Arts" category response rate, from 28.6% to 47%? Did US News perhaps cook their stats by sending fewer such questionnaires!? Amazing if so. 🤔

130

Johan Ugander @jugander.bsky.social · 23/04/2024

🎉 Congratulations to @marionf.bsky.social @kjhealy.bsky.social for their #1 Amazon rating as a new release in Credit Ratings & Repair. Ordinal society gonna ordinal. 🎉

1320

Johan Ugander @jugander.bsky.social · 15/02/2024

Alternative headline: “The crisis in public education could 'supercharge' election disinformation, US tells the BBC”

021

Johan Ugander @jugander.bsky.social · 24/01/2024

Recently spotted this massive email footer from the French consulate in SF, seems to be a big campaign.

010

Johan Ugander @jugander.bsky.social · 22/12/2023

Google scholar leaning in to cumulative advantage. Makes me wonder what their objective function looks like. (For the record, I do follow Daron, but under a second email address that fwds to my main, a practice that dates back to when you could only follow max 100 authors.)

010

Johan Ugander @jugander.bsky.social · 12/10/2023

I would have expected diversity and degree to be more positively correlated than they seem, given how local clustering coeff tends to vary with degree in online social networks (plot below from the 2011 anatomy of FB paper).

100

Johan Ugander @jugander.bsky.social · 12/10/2023

Preparing a lecturing for tomorrow, I noticed that the "LinkedIn weak ties" study (Rajkumar et al., 2022) has a very interesting map of the US showing average degree and "average diversity" = 1-(local clustering coeff) = frac of friends who aren't friends. www.science.org/doi/10.1126/...

120

Johan Ugander @jugander.bsky.social · 06/10/2023

👀 stanford.edu/~jugander/pa...

011

Johan Ugander @jugander.bsky.social · 03/10/2023

This weekend I was reading Rick Ridgeway's autobiography, including this bit about his 1983 expedition to the Vinson Massif. This musing about Lincoln Ellsworth's 1935 flights struck me. It's not clear to me when geographers confidently knew that no peaks in Antarctica were taller than Everest…!

011