In collaboration with DeepMindAI, we introduce the Deep Planning Network (PlaNet), a purely model-based agent that learns a world model from image inputs and successfully leverages it for planning. Learn more and find the source code at ↓ goo.gl/Ckaiof

Google open-sources PlaNet, an AI agent that learns about the world from images venturebeat.com/2019/02/15/goo… via VentureBeat

Combine deep neural nets with a probabilistic model. Construct a Variational Autoencoder (VAE) using TFP Layers and Keras. Here’s how ↓ bit.ly/2TDU5EL

Really excited and honored that Science Museum in London Science Museum is showing our STOP Sign adversarial example on exhibit ("DRIVERLESS: WHO IS IN CONTROL" buff.ly/2SWrwzB) through Oct 2020 - to build autonomous cars, we need to build more resilient learning systems!

Honored to be named on the Female Founders 100 list #FemaleFounders 100 by Inc. ! Excited to be working with a brilliant team, building the platform for privacy-first applications.

#mdpientropy Learnability for the Information Bottleneck mdpi.com/1099-4300/21/1… #learnability #informationbottleneck #representationlearning #conspicuous subset

Great speaking with Tom Simonite WIRED at #wired25 summit about building a new privacy paradigm where users are data owners instead of serfs: buff.ly/2pV0tdK! And amazing stories from fellow speakers Brian Acton Signal benahorowitz.eth a16z Patrick Collison Stripe & others!

Excited to share Director, a practical, general, and interpretable reinforcement learning algorithm for learning hierarchical behaviors from pixels! Director explores and solves long-horizon tasks with very sparse rewards by breaking them down into internal subgoals. Thread 👇

Demolition of the Deauville Hotel in Miami Beach. #Deauville #demolition #MiamiBeach

SayCan has achieved impressive results, but how do its language model plans get grounded to what robots can achieve? The key ingredient is the PI-QT-Opt agent, which learns robust multitask value functions with sim-to-real RL at scale #CoRL2022 piqtopt.github.io. Details🧵

Huge thanks to ~1K in-person attendees & 4K+ who joined livestream to our Future of Decentralization, AI & Computing Summit, hosted by UC Berkeley RDI! Check out full program slides&videos including keynotes from Shafi Goldwasser Michael Jordan Berkeley AI Research: rdi.berkeley.edu/events/decentr…

Google presents A Human-Inspired Reading Agent with Gist Memory of Very Long Contexts paper page: huggingface.co/papers/2402.09… Current Large Language Models (LLMs) are not only limited to some maximum context length, but also are not able to robustly consume long inputs. To address

We propose ReadAgent 📖, a LLM agent that reads and reasons over text up to 20x more than the raw context length. Like humans, it decides where to pause, keeps fuzzy episodic memories of past readings, and looks up detail info as needed. Just by prompting. read-agent.github.io

Proud to work with a great team! Xinyun Chen, Hiroki Furuta, John Canny, Ian Fischer Paper: arxiv.org/abs/2402.09727 Website & Colab Demo: read-agent.github.io read-agent.github.io

Sculptor: the missing UI for Claude Code 🎨 Imagine running 5 Claudes in parallel, safely in containers, while you stay in flow. Then bring their work straight into your IDE to test/edit together. This is how one developer ships like a team. Try it with Sonnet 4.5!