Boris Hejblum
LightNews LightNews
banner
borishejblum.science
Boris Hejblum
@borishejblum.science
47 followers 200 following 0 posts
INSERM researcher in biostatistics & #rstats enthusiast - views are my own
https://borishejblum.science
Posts Replies Media Videos
Reposted by Boris Hejblum
andrew.heiss.phd
My latest attempt at an AI/LLM policy in my intro to social science stats class. Basically two rules:

1. Everything human-facing must be human-generated
2. You are responsible for understanding, verifying, and citing everything an LLM generates

quantf26.classes.andrewheiss.com/syllabus.htm...
Generative AI and LLMs In this class, I have two general rules regarding LLMs: Everything human-facing must be human-generated. You are responsible for understanding, verifying, and citing everything an LLM generates. I’ll explain what these mean below, but first, an important warning! LLMs and learning Phew, I can talk about the relationship between learning and LLMs for hours (see this for a more formal explanation of my thinking). LLMs and generative AI can be useful for statistical programming, but only when you already know what you are doing. They can be dangerous and counterproductive for beginners. Additionally, using LLMs and generative AI to write does not lead to deeper learning. The point of writing is to help crystalize and organize your thinking. Pasting LLM-generated words into an assignment to make it look like you read and understood the content will not help you learn. Pasting LLM-generated code into an assignment and hoping that it works will not help you learn. Rule 1: Everything human-facing must be human-generated Generative AI tools like Gemini, ChatGPT, and Claude can be helpful for generating ideas or topics for your assignments and can even help you find existing research about topics you’re interested in. They are useful for debugging and troubleshooting code. These uses are permitted in this course. If you want to use LLM tools to document your code, clean up your notes, look up error messages, search for research related to your topic, and so on, cool. Do it. That’s fine. It’s unavoidable nowadays anyway, since every R-related Google search will give you a Gemini-generated answer at the top with mostly working R code. Any writing and revisions, however, must be your own. You may not use AI tools to write any of the text you submit to me. AI text adds nothing to my understanding. I have no interest in engaging with it at all. There is nothing more disheartening for me than spending my time grading something that ChatGPT spat out in 10 seconds. I want to see good engagement with the readings. I want to see your thinking process. I want to see you make connections between the readings. I want to see your personal insights. I don’t want to see a bunch of words that look like a human wrote them. That’s not useful for future-you. That’s not useful for me. That’s a waste of time. Thus, anything human-facing (i.e. not stuff done just for yourself, like your own personal notes, research, etc.) must be human-generated (i.e. written by you). You may not use AI tools to write any portion of your assignments. Using AI tools in this way, or failing to disclose the use of AI tools, will be treated as a case of plagiarism and referred to the Honor Council. Rule 2: You are responsible for understanding, verifying, and citing everything an LLM generates These tools are really good at working with code, but again, only if you know what you’re doing. Without guidance and expertise, LLMs love making lots of extraneous, convoluted, and weird code by default. The code ostensibly works, but it’s often strange and unnecessary and uses uncommon syntax and packages. If you use LLMs for help with code, you must understand and verify what’s going on with your code and you must cite where it came from. This means that you need to know what each line is doing. If the LLM uses a function or gives you an argument that you don’t understand or haven’t seen before, figure out why and figure out if it’s necessary. Look at the documentation for the function. Search Google for other examples. Ask the LLM about it, and then add comments to the code explaining what’s going on. The citation doesn’t need to be anything formal (i.e. don’t worry about Chicago or APA guidelines)—it just needs to (1) say which LLM you used, and (2) mention what you asked the LLM. You should do this by using code comments—inside your code chunk, add a # to the beginning of a line so that it’s treated as a comment (or text) instead of actual code. Here’s an example of what this can look like: Here’s an example of what this can look like: # This calculates the average GDP per capita in each region in the dataset, # for all countries after 2015. I couldn't remember how to use group_by() and # summarise() together, so I asked Gemini: # # "I'm using tidyverse. I have a dataset named my_dataset and I'm filtering it # to include all countries after 2015. I want to calculate region averages with # group_by and summarise but cannot remember the syntax" my_dataset |> filter(year > 2015) |> group_by(region) |> # Gemini here included na.rm = TRUE, which omits any rows with missing values # when calculating the average. That's okay and necessary here because # South Sudan is missing some years of GDP data summarise(avg_gdp = mean(gdp_per_cap, na.rm = TRUE)) # Gemini also included an extra ungroup() function at the end, but I don't need # that because there are no groups leftover after summarising here ↑ That’s a lot of extra comments and you won’t always see stuff like that in real life code, but I want to see it here, and future-you will want to see it too.
August 18, 2026 at 1:55 PM
Reposted by Boris Hejblum
rrmsky.bsky.social
The most pressing questions in immunology — why do people respond differently to infection, vaccination or therapy? Why do some develop chronic inflammatory disease while others remain resilient? See high-dimensional public data resources.
@hipcproject.bsky.social @yalemed.bsky.social rdcu.be/fAwms
From data to discovery: a unified framework for human immunology
Nature Immunology - High-throughput, high-dimensional profiling technologies have transformed the ability to generate and test hypotheses of human immunology. Investment in infrastructure has...
www.nature.com
August 18, 2026 at 1:32 PM
Reposted by Boris Hejblum
simonpcouch.com
I've been really impressed with Kimi K3 and GLM 5.2 the last couple weeks. I'm excited to share that we've made the models available in Posit AI with strong privacy guarantees and a steep discount compared to the Claude models currently available in the service!

opensource.posit.co/blog/2026-08...
Kimi K3 and GLM 5.2 are now in Posit AI
A new batch of open weights models in Posit AI show impressive capabilities at a fraction of the price.
opensource.posit.co
August 10, 2026 at 3:09 PM
Reposted by Boris Hejblum
martynplummer.bsky.social
What's new in JAGS 5.0.0 part 1: Initial Values martynplummer.wordpress.com/2026/08/03/w...
August 3, 2026 at 5:15 PM
Reposted by Boris Hejblum
ikashnitsky.phd
🚨Oh boy! Here is a masterpiece worth reading carefully, saving as a bookmark in a closeby drawer, and taking out this weapon to use in comments whenever you see a legacy usage of ollama
🫂Friends Don't Let Fronts l Friends Use Ollama
🔗 sleepingrobots.com/dreams/stop-...
#foss #llms #ollama #llamacpp
April 18, 2026 + Friends Don't Let Friends Use Ollama Ollama gained traction by being the first easy llama.cpp wrapper, then spent years dodging attribution, misleading users, and pivoting to cloud, all while riding VC money earned on someone else's engine. Here's the full history, and why the alternatives are better. Tags: LLMs, local, open-source, opinion
July 7, 2026 at 3:32 AM
Reposted by Boris Hejblum
coatless.bsky.social
Turn a Shiny app into a desktop app your users just double-click.

shinyelectron wraps #rstats or #Python Shiny in Electron and builds a native installer for macOS, Windows, and Linux, with a runtime of your choice (including a fully offline build). Build locally or in CI.

#rshiny #pyshiny
macOS drag-to-install window for demo-r-app-suite: the app icon with an arrow pointing to the Applications folder. The app's About panel showing the r-shinylive (WebR) backend, R 4.6.0, 37 packages, platform wasm32-unknown-emscripten, inside the desktop app. shinyelectron launcher listing three R Shiny apps as cards (Dashboard, Data Explorer, About), with an "Apps → Back to Launcher" menu open. GitHub Actions run "Build demo installers" for shinyelectron 0.2.0, matrix building 73 of 76 jobs across macOS, Windows, and Linux.
July 2, 2026 at 6:54 PM
Reposted by Boris Hejblum
nxn.se
FMs and Perturbation Prediction: Good Embeddings vs. Fancy Architectures - elijahcole.me/blog/2026-03...
FMs and Perturbation Prediction: Good Embeddings vs. Fancy Architectures
elijahcole.me
June 20, 2026 at 10:09 PM
Reposted by Boris Hejblum
posit.co
Announcing roxygen2 8.0.0 for #RStats package developers! 📦

This major release streamlines documentation workflows and includes over 100 bug fixes, S7 support, R6 tooling, a lighter footprint, and modern configuration.

Learn more: opensource.posit.co/blog/2026-05...
roxygen2 8.0.0
roxygen2 8.0.0 is now on CRAN, with first-class support for S7, a bunch of R6 improvements, a new configuration style, and a slimmer dependency footprint.
opensource.posit.co
June 1, 2026 at 6:10 PM
Reposted by Boris Hejblum
cailinmeister.bsky.social
Am I the only one who is slowly modifying their written voice to make it increasingly un-AI sounding?
June 1, 2026 at 7:03 PM
Reposted by Boris Hejblum
edouard-lhomme.bsky.social
🗓️RDV 5 juin 2026 à 12h30
pour 1er webinaire Santé Mondiale @univbordeaux.bsky.social Isped / Univ Senghor sur « Vaccins contre le paludisme : point d'étape et perspectives » 💉🦠🦟
avec Pr Halidou Tinto, chercheur burkinabé d’exception
May 28, 2026 at 9:07 PM
Reposted by Boris Hejblum
mattansb.msbstats.info
I put together a short slidedeck on specifying random effects in mixed models.

This doesn't cover all possible edge cases, but I think this provides some basic best practices for proper generalization and uncertainty in estimation.

Feedback is welcome!

#stats #MLM #HLM #LMM #MEM (did I miss any?)
Specifying Random Effects Structure
Specifying Random Effects Structure for Hypothesis Testing or Estimation of Fixed Effects A Comprehensive Guide to Correct Random Effects Specification Mattan S. Ben-Shachar
docs.google.com
May 23, 2026 at 5:56 PM
Reposted by Boris Hejblum
posit.co
Four engineers behind Posit Assistant broke down how to build an AI agent for data science — reasoning, agentic workflows, modeling guardrails, and chart extraction.

Watch: www.youtube.com/watch?v=A2sa...

Download RStudio: posit.co/download/rst...

#PositAI #RStats
Downloads: RStudio Desktop
Download the open-source RStudio Desktop IDE for Windows, macOS, and Linux. Be more productive with R and Python using our powerful editor, debugger, and plotting tools.
posit.co
April 27, 2026 at 7:35 PM
Reposted by Boris Hejblum
jtrnyc.bsky.social
I made a template and video tutorial for creating a Quarto book rendered with GitHub Actions and published on GitHub pages -- check it out www.youtube.com/watch?v=o1vn... #quarto #rstats
Using GitHub Actions to render and deploy a Quarto Book on GitHub Pages (with code template)
YouTube video by Joyce Robbins
www.youtube.com
April 28, 2026 at 1:28 AM
Reposted by Boris Hejblum
jumpingrivers.com
R 4.6.0 "Because it was There" is set for release on 24 April.

@russhyde.bsky.social writes up the more interesting changes — %notin% in base R, neater character column summaries, and more.

Full breakdown with code: www.jumpingrivers.com/blog/whats-n...

#rstats #datascience #rprogramming
April 21, 2026 at 10:36 AM
Reposted by Boris Hejblum
paulgp.com
First pass at a website that keeps track of the data availability status of various published papers:

paulgp.com/replication-...

Very much in Alpha mode, so data will be updated soon.
April 13, 2026 at 5:57 PM
Reposted by Boris Hejblum
clauswilke.com
Did you see the blog post on the topic I posted today?
blog.genesmindsmachines.com/p/creating-r...
Reproducible pipelines
There was a discussion recently on Bluesky about reproducible data analysis pipelines.
blog.genesmindsmachines.com
March 28, 2026 at 1:50 AM
Reposted by Boris Hejblum
jef.works
As I'm updating my NIH Biosketch for the N-th time this year, I made a Jekyll theme for students to familiarize themselves with this CV structure while building their online presence.

Demo: jefworks.github.io/online-biosk...
Fork to modify: github.com/JEFworks/onl...

#JustAcademicThings
January 22, 2026 at 3:51 PM
Reposted by Boris Hejblum
liomys.mx
Feb. update to the LLM+R guide 💪

8 new packages including:
code review, predictive modeling, speech-to-text, text-to-speech, HuggingFace integration, Gemini CLI companion, a CLI coding agent written in R🤯 ,and more!

available in English 🇱🇷 and Spanish 🇲🇽
luisdva.github.io/llmsr-book/
#rstats
February 16, 2026 at 5:30 PM
Reposted by Boris Hejblum
jessicahullman.bsky.social
I wrote about what it means to keep writing when more language feels like the last thing we need--as a computer scientist, but also as a writer.
Softly, effectively, in the age of AI
On dolphins and disobedience
open.substack.com
February 5, 2026 at 12:39 AM
Reposted by Boris Hejblum
tiffanytimbers.bsky.social
"Ten simple rules for teaching data science": arxiv.org/abs/2602.02874

A new preprint by @minecr.bsky.social and myself. We'd love any feedback!
Ten simple rules for teaching data science
Teaching data science presents unique challenges and opportunities that cannot be fully addressed by simply borrowing pedagogical strategies from its parent disciplines of statistics and computer scie...
arxiv.org
February 4, 2026 at 4:39 PM
Reposted by Boris Hejblum
seva.bsky.social
I just finished a three-year term as an editor at an international relations journal. I began at the start of the LLM era but ended right in the middle of it. Our volume of submissions tripled and our desk reject rate rose to 75%. I have some thoughts.
open.substack.com/pub/hegemon/...
The Age of Academic Slop is Upon Us
what happens when AI automates "normal science"?
open.substack.com
January 13, 2026 at 3:38 PM
Reposted by Boris Hejblum
adamjkucharski.bsky.social
Want to get the data out of a PDF figure? As in, the actual data – not a rough trace-along-the-lines version?

I made an app you might like: adamkucharski.github.io/pdf2plot/

It all started a few years ago... 🧵
January 13, 2026 at 9:12 PM
Reposted by Boris Hejblum
posit.co
Introducing bluffbench, a new tool to evaluate how well LLMs actually see data plots.

When we trick LLMs with secret #RStats transformations, they can miss the visual contradiction.

bluffbench helps us measure this "blind spot" in AI coding agents. Learn more: posit.co/blog/introdu...
When plotting, LLMs see what they expect to see - Posit
Data science agents need to accurately read plots even when the content contradicts their expectations. Our testing shows today's LLMs still struggle here.
posit.co
November 19, 2025 at 4:55 PM
Reposted by Boris Hejblum
jennybryan.bsky.social
Will definitely include this example in my next talk on how to name files!

✅ Full marks for "make it easy to guess what the heck something is, based on it name".
earlymodernjohn.bsky.social John Gallagher @earlymodernjohn.bsky.social · Oct 30
Genuinely delighted to download a PhD thesis from a university repository where the author has neglected to remove the words "BITCH THIS IS YOUR THESIS" from the filename.
October 31, 2025 at 8:46 PM
Reposted by Boris Hejblum
ivelasq3.bsky.social
I loved discussing "Positron for RStudio Users: A Gentle Introduction" with @simisani.bsky.social & R-Ladies Gaborone!

Check out the recording and materials:

📹 www.youtube.com/watch?v=2fOQ...
📝 ivelasq.rbind.io/talk/positro...

I hope this intro is better than the one between my three cats 😹
Positron for RStudio Users: A Gentle Introduction with an image of a cute orange kitten sniffing a hand
October 27, 2025 at 10:20 PM