Sophie Greenwood

A feed about the atproto ecosystem but minus anything about Jerry, which is what we all want

And it comes back with a great design! Semantic similarity, a nice well-designed compliant regex, a set of high likelihood hashtags, lists of users to focus on, and more!

Great news! Building a feed on Graze got way easier today. Our prompt-based feed generator got a massive facelift over the weekend, and can now help you create the perfect feed using plain language (like this feed about ATProto without any mention of @jcsalterego.bsky.social). Come give it a try!

whoa wait Bible x atproto plans? 👀 if you come to atmosphereconf would be so curious to hear more!

amazing, imo arguing over preprints is the original and truest form of academic networking 🫡 glad you like the feed!

I'm hiring a postdoc at @cmu.edu (w/ far.ai & @dgrand.bsky.social + @gordpennycook.bsky.social)!

How do LLMs shape human beliefs — and what do we do about it? AI safety meets behavioral science.

Open to technical and social science backgrounds.

Wouldn't it be nice if events you find posted on @bsky.app could automatically get added to your calendar? They can now. 👇

it's a tough competition but this is my favorite paper from @kennypeng.bsky.social so far-- studying the language model linear representation hypothesis through the lens of compressed sensing 👀

arxiv.org/pdf/2602.11246

New paper! The Linear Representation Hypothesis is a powerful intuition for how language models work, but lacks formalization. We give a mathematical framework in which we can ask and answer a basic question: how many features can be stored under the hypothesis? 🧵 arxiv.org/abs/2602.11246

First Proof To assess the ability of current AI systems to correctly answer research-level mathematics questions, we share a set of ten math questions which have arisen naturally in the research process of the au...

Ten math problems with proofs known the authors. Proofs are encrypted until Feb 13. For all problems, authors claim both AI-based literature searches and zero-shot attempts at proofs failed. If you want to take a crack, you have until next Friday (2/13)!

@mariaa.bsky.social and I have been building a custom client that we'd like to make public in the near-term BUT we want to make sure we've gotten the security right. Anyone with some expertise open to helping us out a bit?

Painting by Claude

Opus 4.6 is very powerful

Our paper “Inferring fine-grained migration patterns across the United States” is now out in @natcomms.nature.com! We released a new, highly granular migration dataset. 1/9

Microsoft Research NYC is hiring a researcher in the space of AI and society!

Fairness in PCA-Based Recommenders

🎙️ I had a great time joining the Data Skeptic podcast to talk about my work on recommender systems

If you're interested in embeddings, aligning group preferences, or music recommendations, check out the episode below 👇

open.spotify.com/episode/6IsP...

What's Next at Bluesky - Bluesky As we head into 2026, we're entering a new phase for the Bluesky app. Last year was about scaling through rapid growth and getting the fundamentals in place. This year is about leaning into what's wor...

Looks like it's a priority this year!

"On following: Our 'who to follow' suggestions need to improve. These are hard to get right, but they're important. High-quality follows lead to a much better experience. Making great connections between people is the fundamental purpose of the app."

The Astrosky Ecosystem: An independent online platform for science communication and social networking While almost everything that astronomers study occurs in the vacuum of space, astronomy itself does not `happen in a vacuum'. Interactions between scientists, as well as outreach to members of the pub...

"While almost everything that astronomers study occurs in the vacuum of space, astronomy itself does not 'happen in a vacuum'.

Interactions between scientists, as well as outreach to members of the public, improve extensively from access to good communication tools."

Paper day! ⤵️

Screenshot showing "The Most Popular Feeds on Bluesky" ranked by total likes. Three feeds are listed: 1) "For You" by spacecowboy (41,101 likes) - a personalized algorithmic feed based on co-likes, 2) "Popular With Friends" by Bluesky (40,669 likes) - shows popular content from followed accounts, and 3) "Discover" by Bluesky (38,496 likes) - displays trending content from personal network.

For You has become the most liked feed: bluefacts.app/feeds/top

Thanks everyone who gave it a like!

Liking a feed may seem meaningless, but I think it does matter to the feed creators. It definitely has been motivating for me.

Consider going over the feeds you have pinned and giving them a like!

Check out our new paper at #AAAI 2026! I’ll be presenting in Singapore at Saturday’s poster session (12–2pm). This is joint work with @shuvoms.bsky.social, @bergerlab.bsky.social, @emmapierson.bsky.social, and @nkgarg.bsky.social. 1/9

If you want to research or even just learn for yourself about "Social Media, LinkedIn, Facebook, Instagram, Youtube, WhatsApp, or Twitter," then yes, unfortunately, it's almost impossible to do that research externally. This is a big reason why Bluesky's open ecosystem is so exciting to me!

Ah good clarification, the bot filter is different from the 10k like filter-- to filter bots we used a list (curated by our collaborator Adi Tsach for another project) that identified profiles as bots via handle + posting activity and filtered out posts from those profiles

What's the point of doing science if you can't tell anyone about it? We are excited to announce @emily.space as another confirmed speaker for #ATScience!

Hello all! 👋

I’m delighted to share a 🚨 new preprint 🚨:

“Active Evaluation of General Agents: Problem Definition and Comparison of Baseline Algorithms”.

A paper thread! 🤩📄🧵 1/N

The blue is all of Bluesky; the orange and green are views on Skygest / likes by Skygest users respectively, so the latter two are definitely shaped by subscribers!

And please let us know your feedback for Paper Skygest – we’re working on building algorithmic recommendations + tools for controlling your feed, and want to do this in a way that works best for you! (7/7)

We also provide implementation details, and end with thoughts on the potential of #atproto for social media research, including feed experiments on Paper Skygest, interfaces for user control, and experiments with novel algorithm designs. (6/7)

Histogram of top arXiv categories posted/seen/liked.

For posts containing arXiv links, we match IDs with data from @arxiv.bsky.social to get categories. Content on Bluesky skews heavily toward computer science, and Skygest users disproportionately are shown and interact with AI-related content (+ cs.CY honorable mention!). Bot posts are removed. (5/7)

Plot showing that the likelihood that a user likes/reposts a post drops steeply with the post's rank.

We also find evidence of strong feed positioning effects on engagement (consistent with intuition, but to our knowledge rarely shown formally in the literature): this plot shows that the likelihood that a user interacts with a post drops off sharply for posts shown lower in the feed. (4/7)

Skygest usage statistics showing daily user / session counts over time.

Since we launched, Skygest has had over 1,000 daily users on average; to date it's served posts over 2M times. 📈

Observationally, Skygest increases engagement with academic content: on average, a user liked 0.5 more posts about papers the week after adopting Skygest than the week before. (3/7)

(If you haven’t yet pinned Paper Skygest, a personalized feed for Bluesky posts about papers, you can check it out here ⤵️) (2/7)

Title + abstract of the preprint

Excited to present a new preprint with @nkgarg.bsky.social: presenting usage statistics and observational findings from Paper Skygest in the first six months of deployment! 🎉📜

arxiv.org/abs/2601.04253

If you know people in Theoretical CS or adjacent who just joined BlueSky (or just did yourself), please tag them or say hi in this thread! #TCSSky (I'll follow them/you, let's strengthen our research community here)