iiinova — where your digital self keeps evolving

Reposted by Hari Subramonyan

Ian Arawjo @ianarawjo.bsky.social · 11/02/2026

🎉 Thrilled to share that our paper "Reporting and Reviewing LLM-Integrated Systems in HCI: Challenges and Considerations" has been conditionally accepted to #CHI2026! A thread 🧵

1105

Reposted by Hari Subramonyan

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

DraftMarks embeds process traces directly into the document using skeuomorphic encodings (e.g., masking tape, smudge marks, eraser crumbs). Instead of external provenance dashboards, the process becomes legible in situ during close reading. This supports what we call process-aware reading.

111

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

In a controlled study (N=70), DraftMarks significantly improved readers’ ability to reconstruct AI collaboration patterns and perceive human effort—without increasing extraneous cognitive load. Process visibility changed how people judged trust, originality, and learning.

100

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

DraftMarks adapts the same underlying trace data to these reading intents.

100

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

Different stakeholders need different traces. In formative studies: • Teachers wanted evidence of student thinking and struggle • Reviewers prioritized intellectual contribution over workflow details • General readers cared about authenticity and voice

100

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

Readers already interpret physical revision marks (cross-outs, tape, marginalia) as signals of effort, revision, and authorship. DraftMarks repurposes this cultural literacy to visualize AI collaboration.

100

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

111

Hari Subramonyan @subramonyam.bsky.social · 28/01/2026

As AI becomes a writing collaborator, readers face a new problem: what is human thought? What did the AI produce? Existing disclosures tell us that AI was used, but not *how* collaboration unfolded. We introduce DraftMarks, an augmented reading interface for human–AI co-writing. #CHI2026

152

Hari Subramonyan @subramonyam.bsky.social · 13/01/2026

Nice Commentary on HCI Education in the Jan issue of Interactions Magazine that mentions our Gulf of Envisoning Work with Maneesh @magrawala.bsky.social Chris Pondoc, Roy Pea, and Colleen Seifert: interactions.acm.org/archive/view...

041

Reposted by Hari Subramonyan

Anne H. Charity Hudley @acharityhudley.bsky.social · 06/01/2026

All righty y'all! The Journal of Black Language and Culture @jblacjournal.bsky.social is now open for submissions! See our website for information on how to submit and reach out if you have questions or ideas as we learn and grow! www.cambridge.org/core/journal...

cambridge.org

Journal of Black Language and Culture | Cambridge Core

Journal of Black Language and Culture - Anne H. Charity Hudley

0116

Hari Subramonyan @subramonyam.bsky.social · 17/10/2025

In this School's In podcast episode, I discuss the intersections of AI, writing, and critical thinking, and reflect on the central role of HCI in designing purpose-built educational tools! Link below:

041

Hari Subramonyan @subramonyam.bsky.social · 29/08/2025

Check out this great synthesis of the tools for thought workshop we organized at #CHI2025 👇

040

Hari Subramonyan @subramonyam.bsky.social · 26/08/2025

📢 AI is changing the way we create, learn, & reason. But is it augmenting human cognition, or dulling it? Please submit your ideas to our #HCI Special Issue: Generative AI in Tools for Thought. Abstract Deadline: Oct 10, 2025 → tinyurl.com/t4tcfp

Screenshot of the special issue website showing title, organizers and links to submit the article. Manuscript deadline is 28th January 2026. Organizers are Srishti Palani, Gonzalo Ramos, Sean Rintel, and Hariharan Subramonmyam

010

Hari Subramonyan @subramonyam.bsky.social · 29/03/2025

Thank you! If possible, would love to chat on zoom, my lab is researching tools for creating interactive learning content.

100

Hari Subramonyan @subramonyam.bsky.social · 29/03/2025

Would love to learn more about what you are building and the process itself.

100

Hari Subramonyan @subramonyam.bsky.social · 28/03/2025

Congratulations @momins.bsky.social for your honorable mention award at #CHI2025

040

Reposted by Hari Subramonyan

Maneesh Agrawala @magrawala.bsky.social · 21/03/2025

Looking forward to giving a connected pair of upcoming talks! Apr 14: I'll present Unpredictable Black Boxes are Terrible Interfaces for the Data Science Institute in Toronto datasciences.utoronto.ca/event/prof-m...

datasciences.utoronto.ca

Unpredictable Black Boxes are Terrible Interfaces - Prof. Maneesh Agrawala - DSI

Join us for the Inequity Speaker Series with Associate Professor Dustin Duncan, from the Department of Epidemiology and Co-Director of the Social and Spatial Epidemiology Unit, Columbia University.

1114

Hari Subramonyan @subramonyam.bsky.social · 14/03/2025

I agree! They should just do that like UIST. In the past I believe CHI only recorded some of the presentations but I could be wrong

110

Hari Subramonyan @subramonyam.bsky.social · 14/03/2025

I feel you! CHI should get rid of the video presentation or atleast make it optional! It’s way too much work.

010

Hari Subramonyan @subramonyam.bsky.social · 07/03/2025

📢 Parents! If you know or have high school students interested in STEM, check out this research opportunity. We are inviting high school students to participate in a study on an AI-supported tool that creates interactive STEM simulations for learning! Details below:

020

Hari Subramonyan @subramonyam.bsky.social · 07/03/2025

If you are a science teacher in the US, please consider participating in our study!! Details below:

011

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Should AI writing tools move beyond chat-based models? What do you think of our layered paradigm? Check out our paper and let us know!! [9/n] @stanfordhai.bsky.social @stanfordgse.bsky.social Paper: arxiv.org/abs/2502.10638 Video Figure: youtu.be/PWr8uEpxOt0

010

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Why layers? Writing isn’t just knowledge telling—it’s knowledge transformation. Instead of rigid back-and-forths, Script&Shift makes writing modular, fluid, and iterative. Layers let you refine ideas, not just rearrange paragraphs, shaping arguments dynamically. [8/n]

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Example use case: Need to build a strong argument? Script&Shift lets you drop your draft into an argument template, structuring it into Claim, Evidence, Warrant & Rebuttal. Tear & Combine sections, refine with Structure Sam, and polish with Feedback Felix [7/n]

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Example use case: A journalist drafting a feature can test multiple ledes, refine sections with Feedback Felix, adjust tone with Tone Tara, and merge the best parts—all without breaking flow. [6/n]

Calling Tone Tara in Script&Shift from the "/" menu and asking it to convert the writing into a poem. Tone Tara creates 2 new layers as options for desired transformation to choose from.

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

For shifting, Script&Shift lets you reshape and refine your writing effortlessly. COMBINE layers to merge ideas, TEAR sections apart to explore new structures and STACK & FOLD for better organization. [5/n]

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

For scripting, Script&Shift gives you AI Writer’s Friends to assist seamlessly. IDEA IVY sparks ideas, DETAIL DANNY expands depth, STRUCTURE SAM organizes, and TONE TARA refines style. [4/n]

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

The two key interactions: 📝 Scripting → Generate, edit, and refine specific parts of a document 📚 Shifting → Rearrange, combine, or test different structural variations in your writing [3/n]

Calling Idea Ivy in Script&Shift using the "/" menu and asking it to generate ideas for potential characters in the story. Idea Ivy provides 5 suggestions.

Comparing two layers by moving them side by side and specifying comparison constraint.

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Our tool Script&Shift reimagines AI-assisted writing with a layered interface, moving beyond rigid chat and inline suggestions. It blends content development & rhetorical strategy, supports non-destructive editing, and enables expressive, iterative writing—all while protecting cognitive flow [2/n].

110

Hari Subramonyan @subramonyam.bsky.social · 26/02/2025

Current AI-assisted writing interfaces are disruptive. They are either too rigid, too complex, or chat-based, forcing writers into an unnatural back-and-forth. Momin @momins.bsky.social developed a new layered interface paradigm that makes AI-assisted writing more intuitive #CHI2025 [1/n] 👇

Script and Shift Interface showing Combining two layers in Script&Shift's spatial interface by overlapping them vertically and supplying merge instructions

Tearing a layer in Script&Shift by selecting the scissor icon in the toolbar and splitting the layer at the desired paragraph

182

Hari Subramonyan @subramonyam.bsky.social · 18/02/2025

I am still waiting for their response

100

Reposted by Hari Subramonyan

Prof Dynarski @dynarski.bsky.social · 14/02/2025

FANTASTIC ICPSR at U Mich is coordinating the archiving of at-risk federal data (all of it?) You can upload data you have & search for data you don’t have www.datalumos.org/datalumos/

9639345

Reposted by Hari Subramonyan

Kevin Feng @kjfeng.me · 16/01/2025

🤔Giving complex tasks to AI agents is easy—getting them to do exactly what you want isn’t. How can human-AI collaboration give us more reliable & steerable agents? 🍫Introducing Cocoa, our new interaction paradigm for balancing human & AI agency in complex human-AI workflows. 🧵

A screenshot of the first page of an academic figure with a list of authors and a high-level schematic figure describing Cocoa, an interactive system for co-planning and co-execution with AI agents.

2387

Hari Subramonyan @subramonyam.bsky.social · 13/01/2025

Excited to co-organize this #CHI2025 workshop on Tools for Thought! 🎉 Join us in exploring how Generative AI can understand, protect, and augment human cognition. Submit your work to help shape this critical conversation! 🔗 Details & Submission: aka.ms/toolsforthou... ⏳ Deadline: Feb 13, 2025 (AoE)

Poster with details on the workshop: Workshop Themes:
The workshop invites a multidisciplinary community to explore questions related to Generative AI (GenAI) and human cognition:

Relevant theories and perspectives for understanding, protecting, and augmenting human thinking with GenAI.
Opportunities and challenges for designing GenAI systems that protect and augment human cognition.
Methods to study and measure the impact of current and novel GenAI systems.
The impact of GenAI on human skills and workflows.
Principles and guidelines for designing GenAI systems to protect and augment cognition.
Organizers:
The workshop is organized by a group of researchers from institutions like Microsoft Research, Harvard University, Carnegie Mellon University, University of Chicago, Tableau Research, University of Toronto, IBM Research, University College London, and Stanford University. Notable names include:

Lev Tanklevitch (Microsoft Research)
Elena Glassman (Harvard University)
Hari Subramonyam (Stanford University)
Goals:
To build a community and develop research frameworks, theories, principles, and systems for understanding and designing tools that use GenAI to augment or protect human cognition.

1202

Reposted by Hari Subramonyan

Martin Wattenberg @wattenberg.bsky.social · 19/11/2024

See the colors of BlueSky, live! www.bewitched.com/demo/rainbow... This little visualization scans incoming posts and draws a stripe every time it finds a color word.

Stripes of various colors, corresponding to color words found in BlueSky posts.

3239398

Reposted by Hari Subramonyan

Alex Kale @kalealex.bsky.social · 18/11/2024

I am recruiting a postdoc! The research emphasis is participatory design for decision-making interfaces. Check out the two project descriptions on my website. people.cs.uchicago.edu/~kalea/prosp... The position is through UChicago’s Data Science Institute. datascience.uchicago.edu/research/pos...

people.cs.uchicago.edu

DCL

These are active projects in the lab where we are seeking to recruit help.

03319

Hari Subramonyan @subramonyam.bsky.social · 17/11/2024

Hi Dan! Can I please be added? Thanks 🙏🏽

110

Hari Subramonyan @subramonyam.bsky.social · 20/03/2024

Check out our paper for details on the cognitive gaps, mental models, and human theory of mind of LLMs and recommendations for designers. Thanks to my amazing coauthors - @magrawala.bsky.social, Roy Pea, Chris Pondoc, Colleen Seifert t.co/QmZEKpwyhZ

t.co

Bridging the Gulf of Envisioning: Cognitive Design Challenges in...

Large language models (LLMs) exhibit dynamic capabilities and appear to comprehend complex and ambiguous natural language prompts. However, calibrating LLM interactions is challenging for...

131

Hari Subramonyan @subramonyam.bsky.social · 20/03/2024

The challenge is that there are no well-defined interface actions or good system mental models of LLMs. Users face three major hurdles with LLMs: knowing what AI can do (capability gap), how to instruct it (instruction gap), and what to expect in return (intentionality gap).

000

Hari Subramonyan @subramonyam.bsky.social · 20/03/2024

This introduces a new kind of gulf, the "Gulf of Envisioning," which refers to the gap between what users want to achieve and how they create prompt instructions that effectively use the abilities and knowledge of large language models to produce good results.

210

Hari Subramonyan @subramonyam.bsky.social · 20/03/2024

But with LLMs, you have an interaction model where (1) operational scope is not restricted to pre-programmed tasks, i.e., no well-defined action space, (2) all kinds of inputs are allowed with an “anything goes” approach, and (3) outputs are probabilistic rather than determinate.

A block diagram showing missing aspects of Norman's mental model for LLM interactions, namely system mental models and action specification, and three pathways annotated on the model.

100

Hari Subramonyan @subramonyam.bsky.social · 20/03/2024

In conventional HCI design, good interfaces minimize the “gulf” between execution and evaluation (aka the Norman Model). They do this through well-designed interface *actions* and *affordances* aligned with users’ system *mental models*.

A block diagram of Norman's seven-stage model, but with two new blocks for cognitive task processes and system mental models inserted just after goals and intentions, respectively.

100