Saujas Vaduguru

People adapt their language to communicate more efficiently over time. How can we make models do this? In our recent work, we trained models in self-play, and found that using the right incentives can make models adapt to communicate efficiently even without human demonstrations.

Simple string manipulation can still break LLMs. In our ACL Findings 2026 paper, PBEBench, even GPT-5 drops below 5% accuracy on long-horizon inductive reasoning tasks. 🧵

New paper: We uncover latent tokens: predicted-but-not-decoded positions that can give diffusion models an advantage over autoregressive models. We show AR models can also be augmented with latent tokens, overcoming known failure modes. arxiv.org/abs/2602.03769

What can LLMs tell us about the nature of spoken communication? 🗣️🤖 From an info-theory perspective, speech has two main channels: words (text) and melody of speech (prosody). We propose a new method using spoken LLMs to quantify how these channels conveys meaning (1/n)

Loving the plan mode in cursor. Most annoying thing with coding agents for me was that they often got small but important details wrong (or at least diff from what I had in mind) and trying to fix them afterwards is a pain. Reviewing the planned implementation before the actual build is so much easier and better. Feels like a successful example of an AI tool leveraging critical human in the loop input.

Tired of evaluating LLMs on made-up problems that look nothing like real tasks? Introducing EDIT-Bench, a code editing benchmark built from in-the-wild user interactions in VSCode. Real-world edits are challenging: 𝗼𝗻𝗹𝘆 𝟭/𝟰𝟬 𝗺𝗼𝗱𝗲𝗹𝘀 𝘀𝗰𝗼𝗿𝗲 > 𝟲𝟬% 𝗽𝗮𝘀𝘀@𝟭.

I'm recruiting my first group of PhD students at TTIC! If you're interested, please apply! If you know people who might be interested, please spread the word! Application deadline is Dec 9, 2025, and there is no application fee: ttic.edu/studentapplica…

@HuaYilun led a cool paper on a different approach that makes clever use of conversations between people to elicit this behavior. x.com/HuaYilun/statu…

More details in the paper at arxiv.org/abs/2510.24023 This was a great opportunity to collaborate with @HuaYilun, @yoavartzi, and @dan_fried!

People adapt their language to communicate more efficiently over time. How can we make models do this? In our recent work, we trained models in self-play, and found that using the right incentives can make models adapt to communicate efficiently even without human demonstrations.

Going to present this at emnlp. Come say hi to me Wednesday Nov5 13:00-14:00 Poster 4003 at Hall C

@evanthebouncy is presenting this at EMNLP next week! Find him in Hall C on Wednesday (Nov 5), 13:00–14:00 to chat about this!

Just look at these multimodal refinement instructions! How would we ground them into reasonable executions? Joint work with @wp_mccarthy @evanthebouncy @judyefan @dan_fried @KarlDD @JustinMatejka Paper: arxiv.org/abs/2504.20294 Code: github.com/AutodeskAILab/…

When we instruct an agent to design something, its first output may not be precisely what we want. Humans collaborating refine their creations iteratively. Can we instruct an agent to refine its output? Is language the best medium for these instructions? We explore this in mrCAD.