Pranav Khadpe

Screenshot of paper title: Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence

AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.

Dear ChatGPT, Am I the Asshole?
While Reddit users might say yes, your favorite LLM probably won’t.
We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

The Fall of an Algorithm: Characterizing the Dynamics Toward Abandonment As more algorithmic systems have come under scrutiny for their potential to inflict societal harms, an increasing number of organizations that hold power over harmful algorithms have chosen (or were r...

@narijohnson.bsky.social and Sanika have a FAccT paper about this (and also compiled cases of abandoned algorithms): arxiv.org/abs/2404.13802

Continue & Persist Letter Continue & Persist Letter

Hi Derek, very happy to hear the project resonated with you! You might also enjoy: continueandpersist.org

This work was done with my mentees Olivia Xu (who worked on the initial deployment) and Anna Gu (who's working on the newer extension), and my advisors
Chinmay and Geoff. And it was made possible by a Google Award for Inclusion Research to Chinmay.

What next? We're working on a CMU-wide deployment of the extension. CMU spread the smiley so I think we've always been big on communicating delight :-)

Finally, who has claim to the appreciation when the work is collaborative?

Contributors’ reactions were influenced by how much they felt they had contributed to the package that was thanked.

Hug Reports were also viewed as a valuable metric of use (ofc value determined by adoption). Today, contributors have little visibility into how their packages are used beyond download numbers. Insights into usage are valuable for raising funding and for project decisions.

But it is hard to be specific when praising.

Contributors ofc preferred notes that were more specific but users often didn't know what to say beyond "this is great". Newer version includes scaffolding for users.
For why praise is less differentiated see

Overall, contributors were surprised and happy to receive the Hug Reports.

In our field study, 18 users interacted with the extension for 3 weeks, resulting in messages of appreciation to 550 contributors, 26 of whom participated in subsequent research.

Hug Reports: Supporting Expression of Appreciation between Users and Contributors of Open Source Software Packages Contributors to open source software packages often describe feeling discouraged by the lack of positive feedback from users. This paper describes a technology probe, Hug Reports, that provides users ...

The extension builds on learnings from our field study of a more rudimentary version. We wanted to get a quick sense of how developers and users feel when sending Hug Reports.

Read about it in our #CSCW2025 paper: arxiv.org/abs/2407.20390

Some highlights below.

Extension: marketplace.visualstudio.com/items?itemNa...

It renders a 🙌 next to Python and JS package imports. Right-click it to send thanks to the package or specific modules.

Thanks are sent in batches to authors of 20 most recent commits of the corresponding repo/file, after manual moderation.

Maintaining open-source packages is a thankless job. Users report issues but stay silent when all works well.

We made a VS Code extension for "Hug" Reports. Try it (link in🧵) and send thanks today!

How do contributors feel receiving Hug Reports? Read our CSCW paper(link in🧵)