Zachary Levonian's Blog — Index
View all my peer-reviewed research on my homepage. The paper that best represents who I am and what I value is “Peer Recommendation Interventions for Health-related Social Support”, which I published in the Computer Supported Cooperative Work conference in 2025.
This page lists my self-published writings, including 79 posts hosted on this blog, 4 hosted on Medium, and a few hosted elsewhere.
Posts
-
Overfitting to train/test splits
Ben Recht writes in a 2023 blog post that train-test splits are shockingly effective for evaluating machine learning models. -
Two approaches to statistical modeling?
Three papers that break down the goals of statistical modeling. -
Destroy 'AI', but keep designing with ML
Ali Alkhatib argues in a blog post that we should destroy AI. I see a few glimmers of hope. -
Generating math hints with large language models
Large language models are pretty good at generating hints for students, given the right constraints. -
Monitoring student safety during generative chats
Ensuring student safety with monitoring dashboards for students' LLM chats. -
"What are the odds?" Stardew Valley edition
What's the probability of having exactly 111,111g in Stardew Valley? -
Using retrieval-augmented generation to improve math question-answering
Humans prefer large language model responses to students' conceptual math questions when created with retrieval-augmented generation and "just the right amount" of prompting guidance. -
Graphing Shakespeare and dramatizing data
Coupette et al.'s paper "All the world’s a (hyper)graph" -
Designing LLM tutors to check for student understanding
What can Doug Lemov teach us about designing LLM tutors? -
“Wishful Mnemonics” in Machine Learning Research
An always-relevant term from researcher Melanie Mitchell. -
AI2 Dolma Most Frequent Words: a derivative dataset of the most common words
A dataset of word counts in Dolma. -
Basics of machine learning with text data: bag-of-words for linear regression
Twitter data workshop presented to undergraduate HCI researchers. -
NLP for Social Computing Workshop
Workshop presented to graduate HCI researchers.
Posts on Medium
-
How would you deal with an ambiguous problem?
Conceptual workflow for scoping Data Science problems. -
Model vs Modeler: Trade-offs designing interactive text classification interfaces
Accessible summary of my IUI 2022 paper. -
Integer Linear Programming with PuLP: Optimizing a DraftKings NFL lineup
Basic introduction to ILP with Python. -
Prototyping a handheld with the Omega2: A complete beginner's guide
Tutorial for making a battery-powered handheld.
Short posts
-
Using Docker with pytest for integration testing
A few packages using Docker with Python's pytest for integration and functional testing. -
Research paper: "Inference-Time Scaling for Generalist Reward Modeling"
DeepSeek paper introducing Self-Principled Critique Tuning for Generative Reward Modeling -
Energy use of machine learning
Link round-up about the energy use of machine learning and large language models. -
Thing precedes theory: motivating the design of new systems
Research through design, unmet needs, and other writing about system-building in human–computer interaction. -
How I write peer reviews
My own process for writing peer reviews. -
Learning webs: Ivan Illich & education technology
Ivan Illich is a controversial writer, but only if you consider destroying all schools controversial. -
Introduction to Human–computer Interaction (HCI)
Resources for gaining familiarity with HCI. -
Research paper: "Peer Recommendation Interventions for Health-related Social Support"
Can recommendation systems help people find health-related peer support online? I wrote a research paper exploring this question. -
Non-linearities in floating point arithmetic
IEEE 754 floating point arithmetic is non-associative. -
Reinforcement Learning Resources
A few resources on reinforcement learning. -
Large language models as actors: Colin Fraser on alignment research
Colin Fraser wrote three Bluesky threads about problems in LLM alignment research. -
Data sharing ethics at Crisis Text Line
Crisis Text Line is a non-profit. -
Stochtree: a library for Stochastic tree ensembles (BART / XBART) for supervised learning and causal inference
Stochtree is a Bayesian Additive Regression Trees (BART) software library. -
Downloading TikTok videos is easy with yt-dlp
TikTok's data export tool combined with yt-dlp makes it easy to download TikTok videos and metadata. -
Prompting large language models
Resources for prompt engineering with large language models. -
AI Safety: is there an existential risk?
A few notes on the existential risk posed by artifical intelligence. -
Which academic journals should have articles on Wikipedia?
An interesting and long-running debate on English Wikipedia inclusion criteria. -
Wrapping Python modules with Wrapt
The Python package Langfuse uses Wrapt to intercept calls to module functions. -
Log-normal distributions in human behavioral data
Log-normal distrubitons appear a lot when measuring the time between human activities. -
Asynchronous code in Python
A collection of links for learning about asynchronous development in Python. -
Resources for building databases and data processing systems
A collection of links for learning about databases. -
DVC: A tool for versioning data
DVC (Data Version Control) is a nifty tool. -
Tutor Co-pilot: real-time scaffolding for math tutors
A project at PLUS using large language models for real-time support. -
Research paper: Ideological differences in the expanse of the moral circle
2019 paper on political ideology and moral circles. -
Research paper: When the implication is not to design (technology)
2011 CHI paper from Eric P.S. Baumer and Six Silberman. -
Can you write a for loop in every programming language on your resume?
An amusing Hacker News comment. -
Research paper: Multimodal, multi-class bias mitigation for predicting speaker confidence
Educational Data Mining (EDM) 2024 paper on predicting perceived speaker confidence. -
Multi-threading for CPU inference in PyTorch
Exploring the performance characteristics of multi-threading for CPU inference with PyTorch. -
Transcribing Zoom recordings with WhisperX on AWS Fargate
A Dockerized transcription and diarization pipeline using WhisperX. -
Machine learning is tarmac
Fred Turner on how information technologies are like tarmac. -
Interpretable machine learning with scoring models
Cynthia Rudin's research on scoring models. -
Annotation, disagreement, and consensus
Collection of research related to annotator disagreement and consensus. -
AWS Copilot CLI makes it easy to run containerized ETL jobs
I missed the release of AWS Copilot CLI back in 2020. It's a useful and opinionated tool. -
Why can you use the pipe operator ('|') in LangChain?
LangChain lets you write chains using vertical pipes. How? By overriding __or__. -
Complexity Sells
Eugene Yan argues that complexity sells. Not during AI winters! -
Research paper: TnT-LLM
Short discussion of the 2024 paper "TnT-LLM: Text Mining at Scale with Large Language Models" -
In-context learning: why does it work?
Why do prompting techniques based on in-context learning improve LLM performance? -
Inference with predicted data
Inference with predicted data, including from text data. -
Research paper: Inconsistent multiple testing corrections
Mark Rubin's position paper on statistical adjustments for multiple comparions. -
Purposeful sampling in qualitative research
Notes on Patton's "Qualitative Research & Evaluation Methods" -
Requirements elicitation for machine learning applications
Requirements elicitation for ML-backed designs can be challenging, through the lens of Qian Yang et al.'s CHI paper. -
Research paper: "Red-Teaming for Generative AI: Silver Bullet or Security Theater?"
A 2024 position paper on red-teaming from Michael Feffer and others at CMU. -
What makes a good short paper at CSCW?
Human-computer interaction and social computing contributions that make for good short papers -
Book: Constructing Methodology for Qualitative Research
2015 book about qualitative research. -
Interesting person: Allie Latimer
Allie Latimer founded Federally Employed Women (FEW). -
Link Dump
Links to stuff I found interesting -
Research paper: Label Sleuth
Shnarch et al.'s EMNLP 2022 paper "Label Sleuth: From Unlabeled Text to a Classifier in a Few Hours" -
Adding references to unreferenced Wikipedia articles
A February 2024 "edit-a-thon" shows the variety in unreferenced Wikipedia articles. -
An interesting two envelopes problem
A probability question posted on Twitter. -
Experimental design: Crowdsourced sources
Crowdsourced experimental design book recommendations. -
"95% of people cannot solve this!"
A fun Quora post about a cruel meme. -
Research paper: Why CSCW applications fail
A classic 1988 CSCW paper. -
Book: Black Software
Charlton D. McIlwain's book Black Software. -
Academic writing: avoid ambiguous antecedents
Avoid ambiguous antecedents, and a collection of academic writing tips from others. -
My debugging process
My process for debugging software. -
Connecting to GCP Cloud SQL from a local Docker container
Connecting to GCP Cloud SQL from a local Docker container using host.docker.internal -
Skrub
Skrub is a cool and useful Python library. -
Research idea: estimate the causal impact of sharing your true opinions on social media
A research/design opportunity for social media communication behaviors. -
Research paper: A Random Sample of YouTube
McGrady et al.'s 2023 paper "Dialing for Videos: A Random Sample of YouTube" -
The CommonLit Ease of Readability (CLEAR) Corpus
A dataset for advancing readability research. -
NeurIPS'23 GAIED Source Round-up
Sources mentioned during the NeurIPS'23 Workshop on Generative AI for Education. -
Converting PDFs to Markdown with Marker
A Python utility for converting PDFs to Markdown. -
Blogroll
Blogs I find to be useful & informative. -
Useful Links
Tutorials and other useful materials -
Who is Zach?
Who is Zachary Levonian as a researcher and engineer? -
Jekyll on GitHub Pages
How this blog is configured.
Other writing stuff
-
Conference reading lists for CHI 2022 and CSCW 2021
Short summaries of recommended human-computer interaction papers. -
Are Bots Ravaging Online Encyclopedias?
[code]
October 2021. Blog post written by Abby Newcomb and Sokona Mangane about our work together during the UMN REU program. (I was a research mentor.) -
A Reflection on Reflective Writing Center Work [pdf]
November 2016. Journal article published in WLN: A Journal of Writing Center Scholarship, authored by Renata Fitzpatrick, Julia Kroll, and myself while I was a Lead Writing Consultant at the Carleton College Writing Center.
The text contents of this site are licensed CC BY 4.0. In other words, feel free to use or reuse them (with attribution to Zach or a link to the original post).
subscribe via RSS