Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities
arXiv
3,400+ cites
Jarred Barber.
Staff Research Engineer at Google DeepMind working on latent reasoning and pretraining for the Gemini models. I build generative models, custom attention kernels, and agent harnesses for math and code reasoning — and I've shipped research the whole way to production, including Google's first generative image product.
4,900+
Citations
10
h-index
4
Filed patents
01 — Selected publications
StyleDrop: Text-to-image synthesis of any style
NeurIPS
310 cites
Muse: Text-to-image generation via masked generative transformers
ICML
880 cites
Improving few-shot classification with unlabeled examples
WACV
95 cites
Sparse Gaussian processes via parametric families of compactly-supported kernels
02 — Things I've built
eigen-QP
A compact quadratic programming solver built on Eigen.
C++
Cat.jl
Composable domain-specific languages with multiple interpretations.
Julia
forbidden-donut
A spinning 3D torus rendered in the terminal.
Rust
ai-artist-gallery
AI artists making 4096² generative code pieces, each in its own voice.
Generative · Read →03 — Get in touch