James Cohan

Excited to share a new paper that aims to narrow the conceptual gap between the idealized notion of Kolmogorov complexity and practical complexity measures for neural networks.

View the full paper here: arxiv.org/abs/2412.16429

Excited to share our work on improving Gemini for learning!

(1/6) 📢 Thrilled to share that our technical report "LearnLM: Improving Gemini for Learning" is out, alongside our #LearnLM model that we released on AI Studio! #google #ai #education #EdTech #Gemini Highlights and links below ⬇️

Excited to share a new paper: “ALTA: Compiler-Based Analysis of Transformers” (w/ @James_Cohan, @jacobeisenstein, @kentonctlee, @JonathanBerant, @toutanova) arxiv.org/abs/2410.18077

“Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach” is now available at arxiv.org/abs/2407.12687  #ICML2024: Irina Jurenka and Markus Kunesch will be demoing the LearnLM-Tutor at the GDM booth on Tues afternoon: deepmind.google/discover/event…

What if everyone, everywhere could have their own personal AI tutor, on any topic? 💡 We’re making learning more engaging and personal with our new family of models, LearnLM. Find out more → dpmd.ai/3wK6fBo #GoogleIO

From Pixels to UI Actions: Learning to Follow Instructions via Graphical User Interfaces Shows that, based on pixel-based pretraining, it is possible for an agent to outperform human crowdworkers on MiniWob++ of GUI-based instruction following tasks arxiv.org/abs/2306.00245

Excited to present Pix2Act! An agent that can interact with GUIs using the same conceptual interface that humans commonly use — via pixel-based screenshots and generic keyboard and mouse actions -- arxiv.org/abs/2306.00245 (1/4)