Sign in

Zach Levonian

@zwlevonian.bsky.social
658 followers 1.4K following 213 posts

Human–computer interaction researcher. PhD from University of Minnesota. Tacoma, WA. Mastodon: zwlevonian@hci.social

PostsRepliesMedia
Zach Levonian @zwlevonian.bsky.social · 03/10/2026
I'll be attending NCME's "Artificial Intelligence in Measurement and Education Conference" next week for the first time! You can follow the conference in the Lea feed: lea.ac/conferences/... #AIMECon2026
lea.ac
Lea - Feed · AIME-Con 2026
Oct 5 – Oct 8, 2026 · Wyndham Grand Pittsburgh Downtown, Pittsburgh, PA, USA · Schedule, papers, and attendees on Lea
100
Reposted by Zach Levonian
conputer dipshit @davidcrespo.bsky.social · 19/09/2026
now this is fun. guest post on Terry Tao's blog by the great Grant Sanderson of 3Blue1Brown arguing that we should "give novel and compelling motivated explanations academic credit similar to what generating new proofs of open problems has had historically."
terrytao.wordpress.com
If math is more than proof, we need to better celebrate the rest of it
[This is a guest post by Grant Sanderson. This blog post was initially written in a different file format and converted using AI. — T.] A sentiment echoing throughout the mathematics communit…
519031
Zach Levonian @zwlevonian.bsky.social · 16/09/2026
Interesting persimmon.humansand.ai/blog/persimm...
persimmon.humansand.ai
Introducing Persimmon | humans&
A model of human behavior, and new ways to measure how faithfully AI can simulate people.
1134
Zach Levonian @zwlevonian.bsky.social · 14/09/2026
voxdev.org/topic/techno...
voxdev.org
Will open source AI keep up for long?
Open source AI is likely further behind closed-source than current benchmarks indicate. If this gap grows, which is certainly possible, it would limit the usefulness of open source AI across a broad s...
010
Zach Levonian @zwlevonian.bsky.social · 13/09/2026
Great thread, mostly because of the phrase “superintelligent Amelia Bedelia”
011
Zach Levonian @zwlevonian.bsky.social · 11/09/2026
I had missed this nice discussion about implementing tutoring interventions. From January, linked in the piece: schooledbymikepetrilli.substack.com/p/how-hard-is-tutoring-to-get-right
schooledbymikepetrilli.substack.com
How hard is tutoring to get right?
SCHOOLED | Wednesday, 1/21/26
020
Zach Levonian @zwlevonian.bsky.social · 10/09/2026
I have officially lost control of my Saved posts reading list on Bluesky. Oldest saved post is now 5 months old, 50+ posts saved.
100
Reposted by Zach Levonian
Matt Barnum @mattbarnum.bsky.social · 30/07/2026
New: There's been a notable and, in many cases, sustained rise in teachers leaving the classroom over the last four to five years, according to new data I collected. The uptick in any one year was modest but the cumulative loss of teachers has been substantial. www.chalkbeat.org/2026/07/30/t...
12312
Reposted by Zach Levonian
Manoel Horta Ribeiro @manoelhortaribeiro.bsky.social · 20/07/2026
Hot take: ACM should just provide torrents called all_papers_{year_month_day}.tar.gz and let anyone download it!
0152
Zach Levonian @zwlevonian.bsky.social · 17/07/2026
Adding "managed $1.9 billion in S3 spend" to my resume.
Screenshot of alert email: Dear AWS Customer,

You requested that we alert you when the actual cost associated with your org-total-monthly budget exceeds $800.00 for the current month. The month actual cost associated with this budget is $1,989,742,178.79. You can find additional details below and by accessing the AWS Budgets dashboard.

A table shows the Alert Threshold as $800. The Actual Amount is $1,989,742,178.79.
130
Zach Levonian @zwlevonian.bsky.social · 17/07/2026
Glad to see the ACM starting a consultation on LLM access to the ACM Digital Library. Here's the feedback I submitted.
Please provide any feedback you would like to share with ACM Leadership below:

I support giving automated tools better pathways to access the ACM DL. As an ACM author, I strongly supported the open-access initiative. Better tooling, metadata, and pathways to access further the mission of the ACM. Practically, I publish with ACM venues because they are an effective way to disseminate my work, and so I expect the ACM to add additional access pathways (such as the HTML view) that make the work more accessible. That includes pathways (potentially paid) for LLM-based access.

One auxiliary note: the protections required to sustain the ACM DL against unauthorized scrapers also have a significant cost for humans. I am a Wikipedia editor, and I regularly attempt to cite ACM publications. However, the automated citation formatting tool that Wikipedia provides to editors (Citoid) uses Zotero translators, which no longer function correctly on ACM DL URLs! This is a real cost to authors; their work will be less likely to be formatted correctly (with full author info, DOI, etc.). That is part of why I support the creation of alternate pathways that enable appropriate access, including to "public goods" like Wikipedia and Semantic Scholar.
020
Zach Levonian @zwlevonian.bsky.social · 16/07/2026
If you work in structural biology, neurodegenerative diseases math, or climate science and use – or avoid! – LLMs in your research, you should talk to the team at RenPhil doing a landscape analysis: aixscience.renphil.org/engagement
aixscience.renphil.org
Airtable | Everyone's app platform
Airtable is a low-code platform for building collaborative apps. Customize your workflow, collaborate, and achieve ambitious outcomes. Get started for free.
030
Zach Levonian @zwlevonian.bsky.social · 05/07/2026
Learned about Spanish Black Legend a few years ago en.wikipedia.org/wiki/Black_L...
en.wikipedia.org
Black Legend - Wikipedia
010
Zach Levonian @zwlevonian.bsky.social · 05/07/2026
Animated a card trick that is secretly the Josephus problem. levon003.github.io/2026/07/04/c...
6 cards as a stack, with the bottom card revealed. The same 6 cards in a ring.
000
Zach Levonian @zwlevonian.bsky.social · 30/06/2026
hmm
The True Believer, patron saint Amanda Askell
000
Zach Levonian @zwlevonian.bsky.social · 29/06/2026
A few notes on using atproto and Citoid/Zotero to enrich my Bluesky likes with link metadata levon003.github.io/2026/06/28/b...
levon003.github.io
My auto-updating Bluesky reading list
A vibe-coded pipeline to extract a reading list from my like history on Bluesky.
010
Zach Levonian @zwlevonian.bsky.social · 29/06/2026
Vibe-coded an auto-updating list of the 2000+ links I've liked on Bluesky. levon003.github.io/bsky-links/
Most-liked domains:

    arxiv.org 223
    substack.com 147
    github.com 79
    simonwillison.net 72
    github.io 71
    doi.org 52
    acm.org 49
    nytimes.com 36
    huggingface.co 33
    youtube.com 33

…and the accounts whose links I've liked the most:

    @simonwillison.net 62
    @davidcrespo.bsky.social 56
    @beenwrekt.bsky.social 31
    @tedunderwood.com 29
    @emollick.bsky.social 28
    @gracekind.net 27
    @dingdingpeng.the100.ci 27
    @timkellogg.me 24
    @mariaa.bsky.social 23
    @lastpositivist.bsky.social 21
020
Zach Levonian @zwlevonian.bsky.social · 15/06/2026
Claude Opus is great at pulling out data from low-res plots. So easy to recreate a figure as an SVG with different choices. levon003.github.io/2026/06/13/h...
levon003.github.io
Research paper: Undergraduate usage of generative AI in spring 2026
A survey of how undergrad students use, trust, and write with GenAI.
010
Zach Levonian @zwlevonian.bsky.social · 14/06/2026
84% of undergrad students report using generative AI at least once a week. Interesting survey.
Frequency of GenAI usage by discipline. Humanities & Arts students have a much higher rate of never using genAI. Amusingly, none of the business/finance students say they never use genAI. The underlying percentages and counts are available at https://levon003.github.io/2026/06/13/henricksen-gallagher-genai-survey.html
010
Zach Levonian @zwlevonian.bsky.social · 13/06/2026
This strikes me as a common mistake among tech futurists. It is relatively easy to tell a story that says (1) our society has been defined by this one scarce resource. (2) But technology is about to make that resource abundant. This changes everything. (3) Just imagine the wonderful possibilities once this new era of abundance has arrived.
They're missing a critical fourth step: (4) With every new abundance, there will be a new scarcity. And how we manage that scarcity will tell us what sort of society we will be living in.
Look toward the scarcity, not the abundance.
14310
Reposted by Zach Levonian
conputer dipshit @davidcrespo.bsky.social · 08/06/2026
this is well put. often AI work is too cooked by the time you hand it off. it‘s hard to figure out how to get your hands into it as a reviewer or collaborator because it’s too smooth, has no obvious seams blog.janestreet.com/i-design-wit...
But there’s a downside: in this workflow, the reviewer is given a fully baked feature. Does that mean they have zero input on the functionality and are just supposed to review the code? Review is not the most fun work—the equivalent in the design world would be getting a detailed wireframe from a PM and being asked to make it look good. I want to make my proposal as clearly and completely as possible, but I still want my engineering teammates to treat it the same way they’d treat a mockup in Figma, as something they and I can iterate on together in design-space
1422
Zach Levonian @zwlevonian.bsky.social · 10/06/2026
Anthropic's Model Naming, Extrapolated (Sam Wilkinson) samwilkinson.io/posts/2026-0...
Model Name 	Description
Aphorism 	One sentence, but it always feels right
Haiku 	Small poem, small bill
Marginalia 	Provides unprompted commentary on your code
Abstract 	Summarizes reasoning it hasn’t done
Sonnet 	Medium poem, medium bill
Diatribe 	Sonnet, but angry
Opus 	Long poem, entire bill
Treatise 	Opus, but citation is left as an exercise for the reader
White Paper 	Treatise, but you have to enter your email to get it
Mythos 	Opus, but scary
Fable 	Mythos, until the question matters
Fable (xhigh) 	Bankruptcy speedrun
Saga 	Fable, but with extra meandering
Saga (Unabridged) 	Includes answers to unrelated questions
Canon 	Refuses to contradict prior answers
Lore 	Interpretation requires a wiki
Cinematic Universe 	Multiple Sagas with a Canon dispatch layer
040
Reposted by Zach Levonian
Mark Riedl @markriedl.bsky.social · 02/06/2026
NeurIPS statement on the use of AI in the position paper track blog.neurips.cc/2026/06/02/a...
blog.neurips.cc
AI-Generated Papers in the NeurIPS 2026 Position Paper Track – NeurIPS Blog
0127
Zach Levonian @zwlevonian.bsky.social · 30/05/2026
Useful round-up. I've been trying out Handy (handy.computer) recently, and it has been pretty reasonable for me.
110
Zach Levonian @zwlevonian.bsky.social · 25/04/2026
"Coding Models Are Doing Too Much" nrehiew.github.io/blog/minimal...
nrehiew.github.io
Coding Models Are Doing Too Much | wh
051
Reposted by Zach Levonian
Kenny Peng @kennypeng.bsky.social · 15/04/2026
🧵 On social media, we feel trapped in “filter bubbles” and “echo chambers.” Often, we see this as a problem with the algorithm. Our new essay argues the problem is the feed interface itself: it constrains our movement—all we can do is scroll. We lay out an alternative vision:
leaflet.pub
From Feeds to Trails
To design the future of social media, rethink the interface before the algorithm.
910124
Zach Levonian @zwlevonian.bsky.social · 16/04/2026
I have 100s of the same bad photo of Mount Rainier. I take a new one every time the mountain is out. All just to delight me.
Mount Rainier, from along the Point Ruston shorelineMount Rainier, from along the Point Ruston shorelineMount Rainier, from along the Point Ruston shorelineMount Rainier, from along the Point Ruston shoreline
030
Zach Levonian @zwlevonian.bsky.social · 04/04/2026
I wrote up a few notes about structuring research narratives. levon003.github.io/2026/04/03/s...
levon003.github.io
Structuring research narratives
How to structure research narratives for talks, papers, and posters
140
Zach Levonian @zwlevonian.bsky.social · 31/03/2026
Renaissance Philanthropy has announced an "AI in early literacy" initiative with the north star goal of halving the number of struggling readers through personalized literacy interventions. www.renaissancephilanthropy.org/insights/ren...
Text: With breakthroughs in AI and other emerging technologies, every young learner has the potential to receive high-quality, personalized literacy interventions at the right time. Through its program focused on early literacy outcomes, the Learning Engineering Virtual Institute (LEVI) will spur the development, scaling, and implementation of AI-driven tools designed to reduce the number of struggling readers by half in the early grades.
000
Zach Levonian @zwlevonian.bsky.social · 29/03/2026
This is something I really worry about. Wikipedia relies on the "network of trust" established by institutional publishers and the editorial oversight they provide. With trust in media publishers plummeting and the rise in independent journalists, how can we establish new trust networks?
261
Reposted by Zach Levonian
Nathan Lambert @natolambert.bsky.social · 10/03/2026
Good post on how to think about honing your skills as an (academic) researcher by Carlini nicholas.carlini.com/writing/2026...
nicholas.carlini.com
How to win a best paper award (or, an opinionated take on how to do important research)
An opinionated perspective on how to do important research that makes a difference (and sometimes win awards).
0193
Zach Levonian @zwlevonian.bsky.social · 04/03/2026
Wikipedia celebrating its 25th anniversary with their "Knowledge is human" campaign. I like the narrative framing they chose.
120
Reposted by Zach Levonian
Simon Willison @simonwillison.net · 23/02/2026
I've published the first two chapters of a new guide to Agentic Engineering Patterns - coding practices and patterns to help get the best results out of coding agents like Claude Code and OpenAI Codex simonwillison.net/2026/Feb/23/...
simonwillison.net
Writing about Agentic Engineering Patterns
I’ve started a new project to collect and document Agentic Engineering Patterns—coding practices and patterns to help get the best results out of this new era of coding agent development …
1319721
Reposted by Zach Levonian
Jessica Hullman @jessicahullman.bsky.social · 23/02/2026
AI makes continuous reproducibility and robustness testing trivial. What happens to science under new levels of scrutiny and stress-testing by default? Some thoughts on how this could play out, informed by watching open science play out over the last decade.
open.substack.com
Living the metascience dream (or nightmare) with AI for science
What happens when we go from replication crisis to robustness extremes?
15719
Zach Levonian @zwlevonian.bsky.social · 21/02/2026
Now immortalized on "List of hoaxes on Wikipedia" en.wikipedia.org/wiki/Wikiped...
Screenshot of table:
Hoax 	Length 	Creation date 	Deletion date 	Size (bytes) 	Links
Jacob Isaacson
Fictitious composer 	17.49 years 	June 29, 2008 	February 19, 2026 	3,722 	Deletion discussion, including admission of hoax by creator
Archived version of the hoax
Rainbow fish (mythology)
Fictitious Hindu myth 	17.49 years 	December 13, 2005 	June 9, 2023 	1,667 	Deletion discussion
Archived version of the hoax
Jay Orpin
Fictitious songwriter and producer; claimed to have worked with famous musical acts including the Backstreet Boys, NSYNC, Bon Jovi, Britney Spears, My Chemical Romance, Fall Out Boy, Linkin Park, and numerous others 	17.48 years 	January 2, 2007 	June 25, 2024 	4,666 	Deletion discussion
Archived version of the hoax
021
Reposted by Zach Levonian
Cat Hicks @grimalkina.bsky.social · 19/02/2026
Pinning everything on the results of a single study was always going to be a volatile way to use evidence, but now it could be very costly indeed. Rely on rigorous areas with deep commitments to clear methodologies, diverse data, and real-world tests of interventions.
041
Zach Levonian @zwlevonian.bsky.social · 19/02/2026
Got to do something cool on Wikipedia yesterday: I uncovered a hoax article that was undiscovered for 17 years! levon003.github.io/2026/02/19/s...
Screenshot of Wikipedia article: Jacob Isaacson (May 5, 1911 – September 8, 1980) was an American composer and musician. Isaacson was most noted for his own Colortone musical notation and his early works within this system. His association with the Fluxus movement was played down by Isaacson, who held European classical tradition in high regard, although his experimental and minimalist compositions drew inevitable comparison.
0115
Reposted by Zach Levonian
Benjamin Mako Hill @mako.cc · 16/02/2026
Cory does a better job of explaining my research than I do. 😅
1104
Zach Levonian @zwlevonian.bsky.social · 11/02/2026
Fun paper. From 2011, but very relevant to the personality design of LLM tutors today.
Title: Rudeness and Rapport: Insults and Learning Gains in Peer Tutoring

Authors:Amy Ogan, Samantha Finkelstein, Erin Walker, Ryan Carlson, Justine Cassell

Abstract. For 20 years, researchers have envisioned artificially intelligent
learning companions that evolve with their students as they grow and learn.
However, while communication theory suggests that positivity decreases over
time in relationships, most tutoring systems designed to build rapport with a
student remain adamantly polite, and may therefore inadvertently distance the
learner from the agent over time. We present an analysis of high school friends
interacting in a peer tutoring environment as a step towards designing agents
that sustain long-term pedagogical relationships with learners. We find that tu-
tees and tutors use different language behaviors: tutees express more playful-
ness and face-threat, while tutors attend more to the task. This face-threat by the
tutee is associated with increased learning gains for their tutor. Additionally, a
small sample of partners who were strangers learned less than friends, and in
these dyads increased face-threat was negatively correlated with learning. Our
findings support the idea that learning companions should gradually move to-
wards playful face-threat as they build relationships with their students.
110
Zach Levonian @zwlevonian.bsky.social · 09/02/2026
tSNE did a pretty good job here!
@zwlevonian, pictured in a cloud of points labeled "Human-Computer Interaction Researchers", adjacent to bubbles for Machine Learning Researchers and AI Scientists, Mathematics Education Leaders, and – strangely – Oregon State Beaver Fans.
000
Zach Levonian @zwlevonian.bsky.social · 03/02/2026
WikiEdu partners with college instructors to have students edit Wikipedia. ~5% of student edits were bad gen AI use, using a Pangram detection system. Biggest problem was mismatch between generated sentence and the associated citation, which is hard to catch. wikiedu.org/blog/2026/01...
wikiedu.org
Generative AI and Wikipedia editing: What we learned in 2025
Like many organizations, Wiki Education has grappled with generative AI, its impacts, opportunities, and threats, for several years. As an organization that runs large-scale programs to bring new e…
000
Reposted by Zach Levonian
Ken Holstein @kenholstein.bsky.social · 02/02/2026
I am recruiting a postdoc in HCI & AI!
 Interested in augmenting human thinking and learning, futures of work & education, participatory AI, collective intelligence, or related topics? Please get in touch! This 2-year position will be based at a top university in Europe (email or DM for more info)
1149
Reposted by Zach Levonian
norvid_studies @norvid-studies.bsky.social · 01/02/2026
--dangerously-skip-understanding
011715
Reposted by Zach Levonian
Armin Ronacher @mitsuhiko.at · 31/01/2026
The future is software writing its own software. Which is why I'm so in love with Pi: a coding agent that can extend itself :) lucumr.pocoo.org/2026/1/31/pi/
lucumr.pocoo.org
Pi: The Minimal Agent Within OpenClaw
A gentle introduction to the Pi coding agent and why I think it’s a glimpse into the future of software.
38315
Zach Levonian @zwlevonian.bsky.social · 24/01/2026
One of my first dates with my future wife was at Glam Doll. I’ve walked by that spot dozens of times going to the gym. Abolish ICE.
000
Zach Levonian @zwlevonian.bsky.social · 15/01/2026
Reasonable take. I like the idea of more CS pedagogy focused on understanding real-world complex codebases. the-learning-agency.com/the-cutting-...
Screenshot of text: If working code no longer reliably signals comprehension, education must optimize for something else: the ability to reason about systems – how components interact, where constraints live, and how changes ripple through a codebase.

For junior developers, this shift is especially important. As AI accelerates output, leveling up now depends less on writing more code and more on understanding existing systems along with their architectures and embedded trade-offs.
1291
Zach Levonian @zwlevonian.bsky.social · 10/01/2026
I missed this last month from @tilmanbayer.bsky.social: "AI finds errors in 90% of October's Featured Articles". Great example of human-in-the-loop LLM use for verifying Wikipedia articles. en.wikipedia.org/wiki/Wikiped...
en.wikipedia.org
Wikipedia:Wikipedia Signpost/2025-12-01/Opinion - Wikipedia
2329
Zach Levonian @zwlevonian.bsky.social · 09/01/2026
Did a quick link roundup for Doll's Verification-Driven Development. Very interesting approach. levon003.github.io/2026/01/07/v...
levon003.github.io
Verification-driven development: using large language models for software development
Doll’s verification-driven development approach.
1244
Reposted by Zach Levonian
Jeffrey P. Bigham @jeffreybigham.com · 06/01/2026
i liked @odannyboy.bsky.social 's breakdown of the four visions of AI futures -- odannyboy.medium.com/the-four-hor...
odannyboy.medium.com
The Four Horsemen of the AIpocalypse
AI is fracturing into at least four different visions of the future.
171
Zach Levonian @zwlevonian.bsky.social · 31/12/2025
I really can’t imagine how different my human–computer interaction PhD would have been if building reasonable-fidelity prototypes took me weeks instead of months…
060