Sanjay Kairam
banner
skairam.bsky.social
Sanjay Kairam
@skairam.bsky.social
Member of Technical Staff @OpenAI | Teaching the machines to learn from humans.

Google Scholar: https://scholar.google.com/citations?user=BH7jpGIAAAAJ&hl=en
Big thanks to @galenweld.bsky.social for leading this work , along with Carl Pearson, @bradspahn.bsky.social, @timalthoff.bsky.social, and @amyxzhang.bsky.social, along with colleagues at Reddit for supporting this research!
September 8, 2025 at 10:34 PM
This paper used a broad set of tools under the hood to study communities: from factor analysis to LIWC-22 to interaction motifs to hierarchical models. Check out the paper to learn more! arxiv.org/pdf/2508.08596
arxiv.org
September 8, 2025 at 10:34 PM
If you design or manage communities, how do you think about encouraging replies, surfacing gratitude, and encouraging inclusive language? What strategies have worked (or not) for you in the past?
September 8, 2025 at 10:34 PM
Style: More expressions of gratitude, feeling, and third-person plural drive SOVC generally. "Netspeak" (e.g. lol, haha, emoji) and a focus on the future are signals of stronger membership & belonging. Is this the case in your favorite communities?
September 8, 2025 at 10:34 PM
Structure: We find that reciprocal replies are great predictors of stronger communities across all three dimensions; reciprocal upvotes drive feelings of membership, while reciprocal downvotes surprisingly don't decrease SOVC.
September 8, 2025 at 10:34 PM
We identify 19 features capturing the style and structure of conversations, which together explain 33% of the variance in SOVC scores, with style capturing more than structure.
September 8, 2025 at 10:34 PM
We find that community experiences (SOVC) on Reddit can reliably be distinguished along three dimensions: Membership & Belonging, Cooperation & Shared Values, and Connection & Influence.
September 8, 2025 at 10:34 PM
Think of a community as an engine. What would you listen for to know if it's running right? The murmur of messages, the rumble of replies, the sparks of support

In this #ICWSM2026 paper, we identify topic-agnostic conversation features that reveal how well a community is vibing.
September 8, 2025 at 10:34 PM
I’m excited about how far we got the program in a short time and am very confident in the Reddit team’s ability to keep making progress on RFR.
November 27, 2024 at 10:02 PM
Thanks!
November 27, 2024 at 1:32 AM
Definitely exciting! Thanks!
November 27, 2024 at 12:07 AM
Thanks -- appreciate the kind words!
November 26, 2024 at 11:28 PM
Thanks!
November 26, 2024 at 11:28 PM
Hey Jeff -- thanks! I've somehow looped all the way back around to our work on "crowd parting"
November 26, 2024 at 10:42 PM
Same!!!
November 26, 2024 at 10:27 PM
Thanks so much! Great to see you in Costa Rica @acm-cscw.bsky.social!
November 26, 2024 at 10:22 PM
Thanks and definitely not! It's been an exceptionally busy couple of weeks but I'll be back posting soon!
November 26, 2024 at 10:22 PM
Thanks!
November 26, 2024 at 9:58 PM
Thanks! Sadly this means fewer collab opportunities in the near future, but hopefully not forever!
November 26, 2024 at 9:58 PM
This new role provides an opportunity to jump into the deep end of understanding how to ensure that these systems perform in ways that align with users' expectations and objectives. To my new team at OpenAI: I'm excited to dive into this next chapter together!
November 26, 2024 at 9:31 PM