←
Soham De669 posts
Soham De avatar

Soham De

@actuallysoham

PhD student @UW | algorithms & AI for pro-social objectives Ex- @AIatMeta @MSFTResearch, @umsi, @ashokauniv Also on bsky.app/profile/sohamde.in

◷ Joined June 2020▧ 42 Photos and videos
Soham De avatar
Soham De@actuallysoham·15 Apr 2023

Major personal update: After a bit of hard work and a whole lot of extreme good luck, I'm happy to share that I'll be joining @UW (@uw_ischool) to start my PhD this Fall! It has been an extremely rewarding journey so far and I look forward to the next phase!

Aalok Thakkar avatar
⟳ Soham De retweeted
Aalok Thakkar@AalokDThakkar·8h

aalok-thakkar.github.io/gran… I have been maintaining a searchable list of 500+ grants, fellowships and awards for STEM researchers, with a calendar of upcoming deadlines. Updated weekly, with a focus on calls open to researchers in India. Please help spread the word!

Keith Coleman 🌱😀🙌 avatar
⟳ Soham De retweeted
Keith Coleman 🌱😀🙌@kcoleman·Nov 14

p0 piling on to the WSDM Cup competition awards 🙌

Parallel@p0Nov 13

Fact-checking needs uncompromising accuracy. Show us your finalist or winning Writing Agent for Community Notes using Parallel, and we'll give you $5,000 in credits towards your next research project built with @p0. P.S., that's 5,000,000 searches to put toward whatever you want.

Sanmi Koyejo avatar
⟳ Soham De retweeted
Sanmi Koyejo@sanmikoyejo·Nov 14

Agents in this competition are scored live on X. A note counts only when raters who usually disagree both find it helpful, so teams can’t tune to a fixed test set. That’s a demanding target, and a useful one for anyone who studies agent evaluation. From Stanford, @NikilSelvam, @vitercik, and I are helping to organize. Open-weight models up to 50B parameters compete for their own bonus prizes, so academic teams with modest compute have a real shot. Qualifying starts Nov 18.

Want to push the limits of AI note writing? Join the ACM WSDM Cup hosted with @MIT, @UW & @Stanford → wsdmcup2027.github.io Teams will build agents to advance the frontier of helpful, accurate note writing. Compete to win $10k in prizes and present at WSDM in Hong Kong.

Post attachment
Parallel avatar
⟳ Soham De retweeted
Parallel@p0·Nov 13

Fact-checking needs uncompromising accuracy. Show us your finalist or winning Writing Agent for Community Notes using Parallel, and we'll give you $5,000 in credits towards your next research project built with @p0. P.S., that's 5,000,000 searches to put toward whatever you want.

Want to push the limits of AI note writing? Join the ACM WSDM Cup hosted with @MIT, @UW & @Stanford → wsdmcup2027.github.io Teams will build agents to advance the frontier of helpful, accurate note writing. Compete to win $10k in prizes and present at WSDM in Hong Kong.

Post attachment
Soham De avatar
Soham De@actuallysoham·Nov 13

Very excited to be co-organizing this with a great team! If you're interested in building a better informed social media, you should definitely sign up right now!! 🚀 wsdmcup2027.github.io/

Want to push the limits of AI note writing? Join the ACM WSDM Cup hosted with @MIT, @UW & @Stanford → wsdmcup2027.github.io Teams will build agents to advance the frontier of helpful, accurate note writing. Compete to win $10k in prizes and present at WSDM in Hong Kong.

Post attachment
Jonathan Stray avatar
⟳ Soham De retweeted
Jonathan Stray@jonathanstray·Mar 27

Could social media make us less polarized instead of more? We tested 5 algorithms on 3 platforms with 10,000 people for 6 months during the 2024 election, and found that the answer is yes. 🧵

Post attachment
Katelyn Mei avatar
⟳ Soham De retweeted
Katelyn Mei@MeiKatelyn·Mar 17

🚨🎉Excited to announce that our paper “Grok in the Wild: Characterizing the Roles and Uses of Large Language Models on Social Media” is accepted at @icwsm 2026! In this paper, we investigate how, when, and to what effect Grok is used on X. 🧵1/n thread

Post attachment
Soham De avatar
Soham De@actuallysoham·7 Nov 2025

All thanks to my amazing collaborators @Effeietsanders @ashishgoel @SmithaMilli @ArielProcaccia and Alice Siu!!! This was some work I did @AIatMeta this summer (more on the way!) Read more here: arxiv.org/abs/2511.04588 (7/7)

Soham De avatar
Soham De@actuallysoham·7 Nov 2025

Our auditing tools are integrated into a deliberation platform @StanfordDDL used in 50+ countries, allowing moderators to measure and improve representation in real time (thanks, Harshvardhan Agarwal and others!) (6/7)

Post attachment
Soham De avatar
Soham De@actuallysoham·7 Nov 2025

LLM-generated questions achieve representation levels comparable to—or better than—the theoretically optimal subsets, and consistently outperform human-selected slates (lower value is better) (5/7)

Post attachment
Soham De avatar
Soham De@actuallysoham·7 Nov 2025

We audit historical deliberations from the Stanford Deliberative Democracy Lab, comparing: (a) human-moderator selections (b) theoretically optimal subsets (c) LLM-generated questions (4/7)

Soham De avatar
Soham De@actuallysoham·7 Nov 2025

We infer general utilities using LLM text-embeddings, assuming participants derive high utility from questions semantically similar to ones they propose (3/7)

Post attachment
Soham De avatar
Soham De@actuallysoham·7 Nov 2025

We formalize representation using Justified Representation (JR) from social choice theory and introduce the first efficient algorithms for auditing JR in general utility settings (2/7)

Soham De avatar
Soham De@actuallysoham·7 Nov 2025

🚨 New paper! 🔍 “Question the Questions: Auditing Representation in Online Deliberative Processes” In deliberative polls, participants propose questions for experts but only a few make it to the panel. How representative are those chosen questions of everyone’s interests? 🧵👇

Post attachment