Jenn Grannen

@jenngrannen

@StanfordAILab PhD, previously @WeAreHRT, @ToyotaResearch, @berkeley_ai. I can teach your robot new tricks.

919Followers

296Following

80Posts total

105.6KViews on collected posts

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

94.8K26 Nov

2.1K

1.4K

1.7K

1.1K

930

1.3K

1.2K

94827 Nov

Last 9 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.62%26 Nov

0.52%

0.84%

0.92%

1.31%

1.08%

1.28%

1.37%

0.84%27 Nov

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes64.2%580 in total

Reposts8.2%74 in total

Replies2.7%24 in total

Quotes2.0%18 in total

Bookmarks23.0%208 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

@jenngrannen Always a pleasure reading your papers Jen! 948 views · 7 likes · 0 reposts · 1 replies 27 Nov 2025 · Open on X →

Had lots of fun working on this one with @michelllepan, Kenneth Llontop, @hocherie1, @mazrk7, @leto__jean, & @DorsaSadigh ! Big thanks also to the Stanford East Asia Library for letting us run around :) Links: 🔗 https://t.co/r2ZXyeKaIf 📄 https://t.co/MvGp9glwB5 🧵8/8 1.2K views · 17 likes · 0 reposts · 0 replies 26 Nov 2025 · Open on X →

Scanford is our first step toward continual, in-the-wild model adaptation. We’re excited for others to build flywheels in new domains, from libraries to hospitals to grocery stores. 📚🏥🛒 🧵7/8 1.3K views · 16 likes · 0 reposts · 1 replies 26 Nov 2025 · Open on X →

The result is a Robot-Powered Data Flywheel, a framework where robots: 1️⃣ Perform real-world tasks 2️⃣ Collect messy, real data missing from internet corpora 3️⃣ Fine-tune foundation models to redeploy stronger Each cycle narrows the gap between models and messy reality. 🧵6/ 930 views · 9 likes · 0 reposts · 1 replies 26 Nov 2025 · Open on X →

This real-world data transformed VLM performance: 📈 Book ID accuracy: 32.0% → 71.8% 🌏 English OCR: 24.8% → 46.6% 🈶 Chinese OCR: 30.8% → 38.0% Scanford doesn’t just adapt to its environment, it also expands the VLM’s general multilingual OCR capabilities. 🧵5/8 https://t.co/s4F 1.1K views · 13 likes · 1 reposts · 1 replies 26 Nov 2025 · Open on X →

We deployed Scanford for 2 weeks across 2,103 shelves in the East Asia Library. It autonomously captured multilingual, faded, and occluded book spines – the exact kind of data missing from internet-scale corpora. In the process, it saved 18.7 hours of librarian labor🕒 🧵4/8 http 1.7K views · 14 likes · 0 reposts · 1 replies 26 Nov 2025 · Open on X →

Our idea💡: Robots don’t just consume foundation models, they can power them. Scanford closes the loop: 📸 Collects real-world data 🧹 Curates it automatically 🧠 Fine-tunes its own VLM 🚀 Deploys again, better each time 🧵3/8 1.4K views · 11 likes · 0 reposts · 1 replies 26 Nov 2025 · Open on X →

Pretrained VLMs, while powerful, are trained on clean internet data. In the real world? Text is occluded, labels are faded, and at the East Asia Library, most books aren’t English! These domain gaps break even the best VLMs. Scanford closes that gap through deployment. 🧵2/8 htt 2.1K views · 10 likes · 0 reposts · 1 replies 26 Nov 2025 · Open on X →

Meet Scanford 📚🤖: a robot that improves foundation models by doing useful work in the wild. Deployed for 2 weeks in the Stanford East Asia Library, Scanford scans books, helps librarians, and continually improves the VLM it relies on. 🔗 https://t.co/r2ZXyeKaIf 🧵1/8 https://t.c 94.8K views · 483 likes · 73 reposts · 17 replies 26 Nov 2025 · Open on X →

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Latest posts

Similar accounts