Context Lab

🚨 HyperTools 1.1 is out! 🚨 This is our group's Python library for working with and visualizing high-D data. v1.0 was a ground-up rewrite; 1.1 fixes lots of bugs and adds some fun/useful new features and better data handling. pip install --upgrade hypertools Read on for a tour! 🧑‍💻

hypertools.readthedocs.ioGallery of Examples - hypertools

We've also added an entirely new "prediction" module, support for hierarchical data (via MultiIndex DataFrames), support for (deformable!) surface meshes, density and volumetric plotting, and more! Check out the gallery of examples to learn more! hypertools.readthedocs.io/en/latest/au...

hypertools.readthedocs.ioThe canonical pipeline order - hypertools

If you're new to HyperTools, welcome! You may want to check out our "canonical data processing pipeline" (hypertools.readthedocs.io/en/latest/pi...) or our tutorials (hypertools.readthedocs.io/en/latest/tu...) to get oriented. Looking forward to seeing what you build!

🌍 20 cities, 1645 months of temperatures = one trajectory. We also make it simple to show complex patterns over time. Here we're showing how temperatures across 20 cities around the globe change from month to month, over 138 years (1875--2013).

We've also added "morph" animations that can interpolate between point clouds or meshes. hyp.plot(shapes, '.', color='k', animate='morph', title=names) Here's a fun little example with some of our new built-in shapes from a "point cloud zoo" 🐰🧊🔮🫖🍾

🎨 hypertools speaks text! Give hyp.plot a list of strings and it embeds them, projects into 2D or 3D, plots the result: hyp.plot(texts, '.') Here, descriptions of 5 paintings becomes their own clouds, drawn in colors pulled from the actual canvases (also showing off our new palette tools!).

🗺️ order='serial' reveals a dataset one piece at a time. Here's an example from Alice in Wonderland's Mad Tea-Party. Individual conversation segments reflect the speaker (color) and what they're talking about (position and shape). If you do research on conversations, this is a fun one!

hypertools.readthedocs.iohypertools

Docs: hypertools.readthedocs.io Code: github.com/ContextLab/hypertools

In our latest study, led by @paxt0n4.bsky.social, we build a geometric framework to ask how our memories change with the passage of time ⏰ Title: Delayed recountings preserve but simplify the semantic geometry of earlier recountings Preprint: osf.io/mhxtd_v1 Code/data: github.com/ContextLab/m... 🧵

osf.ioOSF

If there are "extra waypoints" in someone's recounting --- i.e., events inferable from their neighbors --- then we can simplify our recountings by dropping them, without losing much of the overall structure of the narrative!

This was a multi-year effort by a fantastic group of wonderful scientists from our team, including @paxt0n.bsky.social, Alishba Tahir, Jennifer Xu, and Soo Hwan Park. Check out our paper to learn more, and we hope you'll play around with our code and data too!

Our work suggests that when we recount our experiences to other people, what we may be doing is "leading" them on a journey through a semantic representation space (approximated here using text embeddings).

But just like navigating through the woods using trail markers, not all markers are equally useful. It's necessary to put in a marker at ambiguous locations (forks in the road, overgrown parts of the trail). But markers at obvious locations (or GPS "continue straight!" instructions) are less useful.

Second, delayed recountings looked like smoothed/simplified versions of the immediate recountings. We used the RDP algorithm, originally developed for cartographic simplification, to ask how much each "event" in people's recountings contributed to the overall trajectory shape.

Events whose removals (from the immediate recountings' trajectories) least changed the overall trajectory shapes were the most likely to be omitted in people's delayed recountings.

We used an approach we developed in our prior work (rdcu.be/cpMwZ) to project the episode, and participants' immediate and delayed recountings of it, into text embedding space. The result is a "trajectory" through the embedding space for the episode and each recounting.

rdcu.berdcu.be

We found two interesting things. First, people's second (delayed) recountings were more similar to their own first recountings than to the original episode! Distortions of the original experience (of watching the episode), as reflected in the initial recounting, carry forward to future recountings.

osf.ioOSF

We had participants watch a television episode, then recount it (immediately), and then recount it again after a week-long delay.

nsf.govAward Details - NSF Award SearchFind award details and explore award abstracts and publications.

12 Aug

Super excited to announce our next big project: an NSF-funded collaboration with the @exploratorium.bsky.social to build museum exhibits that adapt on-the-fly to visitor interests, goals, and background knowledge! 🧑‍🔬🩻🔬🥼🧪🧬🤓🤖🚀 www.nsf.gov/awardsearch/...

biorxiv.orgGlobal topology of brain-wide co-fluctuations links task states, personality, and behavioral symptom dimensionsFunctional connectivity in network neuroscience is traditionally characterized using time-averaged correlations between brain regions. While these summaries capture stable large-scale organization, they do not fully reflect the temporal structure of moment-to-moment interactions. Here, we investigate how the order of interaction used to represent brain dynamics shapes the organization recovered from neural data. We compare three interaction representations of fMRI dynamics: regional activation (node time series), pairwise co-fluctuations (edge time series), and higher-order triplet interactions (triangle time series); within a common topological framework using Mapper from topological data analysis (TDA). Across task and resting-state data, Mapper representations derived from pairwise co-fluctuations more distinctly segregate task conditions than activation-based or higher-order representations. This organization reflects structured coordination patterns beyond activation polarity and is driven by high-amplitude interaction events. Beyond task states, modularity quality computed across all Mapper representations is highest for edge time series and selectively associated with stable individual differences: higher modularity relates to higher conscientiousness and lower internalizing and externalizing symptom dimensions. Together, these findings suggest that behaviorally relevant information is reflected in the topology of moment-to-moment brain interactions. Topological analysis of interaction-level dynamics therefore provides a complementary and interpretable framework for linking large-scale neural coordination to cognition, personality, and mental health. ### Competing Interest Statement The authors have declared no competing interest. National Institute of Mental Health, https://ror.org/04xeg9z08, MH127608 Stanford Maternal and Child Health Research Institute, https://ror.org/00yt0ea73, Faculty Scholar Award

6 May

Check out our latest collab with Manish Saggar, @richardfbetzel.bsky.social, et al. (lead: Chunyin Siu)! We use TDA to show that 2nd-order (edge) brain interactions more distinctly segregate task connections than other kinds of network patterns. Paper/code/data: www.biorxiv.org/content/10.6...

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

Curious what a representation of "everything" you know might look like? Wonder how you might fill it in? Check out our demo and paper (led by @paxt0n4.bsky.social and now out in @natcomms.nature.com ), or read on to learn more! Demo: context-lab.com/mapper/ Paper: www.doi.org/10.1038/s414...

youtube.com[MIND 2019] Jeremy Manning: Can we improve real-world learning using scalable AI teachers?YouTube video by MIND Summer School

Where we're going next is using these maps as the core representations of AI tutoring systems. By accurately modeling landscape of someone's knowledge, we think we can use that representation to teach people more effectively: www.youtube.com/watch?v=FMmP...

Great human tutors intuitively build up an understanding of what their students know, and leverage that in their teaching. The knowledge models we've built can give AI tutors the power to do something similar by explicitly modeling (and interrogating) a personalized model of the learner's knowledge.

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

If you're curious what *your* map looks like, you can map out your own knowledge using our web demo: context-lab.com/mapper/. It usually takes around 30 or so questions to converge on a stable representaiton. It can be a highly revealing and humbling thing to learn what you know!

We've also put together other tools for probing and visualizing these maps. Here is a fun project that uses Blender to create synthwave-style 3D renders of knowledge maps: github.com/ContextLab/k...

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

Eventually, for concepts that are too far apart, predictive accuracy falls to "chance" and the model ends up predicting that your chances of answering a question correctly are equal to whatever your overall average performance is, across all questions (regardless of content).

We also found that each person has a unique knowledge map, which can be revealed by asking just a few multiple-choice questions. We're excited about what this unlocks: imagine matching up study partners, conversation partners, or even social relationships by studying how people's maps align!

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

So these tests showed that the maps we get out using this approach reflect people's actual conceptual understanding of real course content! Next, we looked at the geometric properties of the maps. We found a bunch of interesting things.

One finding is that knowledge is fundamentally *smooth*. Given that you've answered a question correctly, the probability of answering another question correctly falls off smoothly with distance. (Same with *incorrect* responses.)

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

Harder is: construct a map using domain X and predict performance on questions from domain Y. This tests the model's ability to *generalize* to unseen, unrelated content.

And even harder: construct a map using all but one question about domain X, and predict performance on the held-out domain X question. This is a highly conservative test of the model's *specificity*, because it requires the predictive model to distinguish between very similar concepts.

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

First, and most importantly: given that someone's knowledge map shows that they have X level of knowledge about some concept, does that correspond to their actual knowledge? We show that the predicted knowledge track closely with people's chances of correctly answering held-out quiz questions.

We tested this in a few different ways, and all of the tests indicated that the maps were predictive of quiz performance! The simplest test is: given a map constructed from all but one question, can we predict how you'll answer that last question?

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

We had our participants answer multiple-choice quiz questions before and after watching a series of @khanacademy.org lectures.

We developed a process for efficiently building a "map" of one person's knowledge using responses to short multiple-choice quizzes. In our paper, we tested two things: (1) the quality of those maps and (2) their geometric properties.

context-lab.comKnowledge MapperAn interactive tool that maps out everything you know. Answer questions and watch your personalized knowledge map take shape.

We use text embeddings to define a coordinate system for human knowledge. Any text (video transcripts, quiz questions, etc.) can be localized in the embedding space, which lets us capture the idea that knowing about one concept also increases the chances that you know about other related concepts.

cdl-scheduler.readthedocs.ioCDL Scheduler documentation

11 Mar

Do you use @youcanbookme.bsky.social or @calendly-official.bsky.social for scheduling? We were frustrated with recent increases in pricing, so we tapped @anthropic.com's Claude to help! You can easily use it yourself: 100% free to set up and host: cdl-scheduler.readthedocs.io/en/latest/

11 Mar

Curious how it works? We use @github.com pages to host a static frontend site, along with @developers.google.com apps scripts as a serverless backend. GitHub actions do scheduled maintenance and Google Calendar, Gmail, and Sheets provide the scheduling, notification, and database infrastructure.

11 Mar

And if you actually want to see the diagram, here you go-- there was some weird blurring in the previous message. More info: cdl-scheduler.readthedocs.io/en/latest/ar...

context-lab.comcontext-lab.com

16 Jan

With our v1.0 release, dream-stream now comes in Android 🤖! context-lab.com/dream-stream...

Jeremy Manning@jeremyrmanning.bsky.social

I made a quirky little web app to help guide your lucid dreams: context-lab.com/dream-stream/ It's kind of like a "netflix" or "spotify" for lucid dreaming-- you select different narratives to form a playlist, and then it uses your device's microphone to start playing when it detects you're in REM.

context-lab.comcontext-lab.com

Context Lab reposted

Jeremy Manning

@jeremyrmanning.bsky.social

11 Jan

context-lab.comcontext-lab.com

Context Lab reposted

Jeremy Manning

@jeremyrmanning.bsky.social

7 Jan

Excited to be teaching a new undergraduate course on Models of Language and Conversation this term! Check it out here: context-lab.com/llm-course/ I've added lots of fun interactive demos of chatbots and NLP techniques that let students dig into the approaches.

context-lab.comModels of Language and Communication - PSYC 51.17Course materials for PSYC 51.17: Language Models from Scratch - Dartmouth College

github.comGitHub - ContextLab/leetcode-solutions: Leetcode discussions, brainstorming, musings, and solutionsLeetcode discussions, brainstorming, musings, and solutions - ContextLab/leetcode-solutions

6 Nov 2025

Remember when grinding leetcode was still a thing? If you'd like to hone your coding skills, or even just return to that simpler time for nostalgia's sake, you might enjoy this project from our group: github.com/ContextLab/l... Happy hacking! 👩‍💻

28 Oct 2025

🚨 New preprint alert! We use trained-from-scratch GPT-2 models to characterize & capture the unique writing styles of individual authors. We also develop a new LLM-based relative stylometric measure. Paper: arxiv.org/abs/2510.21958 Code/data: github.com/ContextLab/l... 🤗: huggingface.co/contextlab

data-wrangler.readthedocs.orgDataWrangler — datawrangler 0.4.0 documentation

14 June 2025

🤠 New release announcement for our datawrangler package! Try it using: pip install --upgrade pydata-wrangler Lots of awesome performance improvements (including native polars support!), simplified API, support for @hf.co text embeddings, etc. More info here: data-wrangler.readthedocs.org

@jeremyrmanning.bsky.social

25 Apr 2025

We're super excited to announce that we've officially convinced @cgonciulea.bsky.social to join our rag-tag (but VERY classy) team of science nerds this fall as a @dartmouthpbs.bsky.social PhD student 🎉🥳🤓🧠🧑‍🔬🎓!!

Context Lab reposted

Jeremy Manning

13 Feb 2025

Just finished a draft of my Models of Memory (grad) course that I'm teaching this spring! Please share/borrow/re-use/follow along as desired, and if you have feedback or suggestions I'd really love to hear (especially while I can still change it)! All materials are here: github.com/ContextLab/m...

github.comGitHub - ContextLab/memory-models-course: Dartmouth graduate course (PSYC 133) on computational memory modelsDartmouth graduate course (PSYC 133) on computational memory models - ContextLab/memory-models-course

Jeremy Manning@jeremyrmanning.bsky.social

Memory-sky: I'm going to be teaching a new (open!) grad course in the spring, where we read about and implement some memory models (and ideally for them to some small datasets). Looking for suggestions to include! I could go "classic" (Hopfield nets, SAM, TCM, etc.), or "modern" with newer stuff 🤔

26 Nov 2024

Re-sharing from Xitter-- new paper out in @pnas.org (lead: Lucy Owen)! We show how the informativeness and compressibility of brain activity patterns change under different levels of cognitive engagement: www.pnas.org/doi/10.1073/... Lots of neat stuff in there!

pnas.orgpnas.org

nature.comTemporal asymmetries in inferring unobserved past and future events - Nature CommunicationsMemories of our past experiences enable us to reconstruct (in the present) our past better than our future. Here, the authors show that we are also able to better reconstruct the pasts (versus futures...

19 Nov 2024

Excited to have this out in @natureportfolio.bsky.social! We found that real & fictional people communicate roughly 1.5x more about the past than the future. In turn, this influences the interferences we make about past & future events in *other* people's lives. www.nature.com/articles/s41...

Context Lab@contextlab.bsky.social

🚨 Updated preprint (lead: @xxming.bsky.social) on how inferences about unobserved past/future events are informed by conversations! osf.io/yp2qu/ 🚨 This update includes a new pre-registered replication + meta analysis. We think we've found some neat fundamental properties of human conversation! 🧪🧠

osf.ioosf.io