engblogs

Simulating matter on the quantum scale with AI to design new materials and solve major challenges of the 21st Century

The blogpost discusses the performance improvements in Chrome on Windows by utilizing the concept of Native Window Occlusion, resulting in faster startup and fewer crashes.

Simulating electrons in order to design new materials for clean electricity and high temperature superconductors.

Creating Interactive Agents with Imitation Learning: Building a multimodal interactive agent that engages in extended and surprising physical and linguistic interactions with humans using imitation learning.

Improving language models by retrieving from trillions of tokens

Language modelling at scale: Gopher, ethical considerations, and retrieval

Imitation learning and self-supervised learning enable the creation of a multimodal interactive agent, MIA, that successfully interacts with non-adversarial humans 75% of the time, with improved performance using hierarchical action selection.

Exploring RETRO: Augmenting transformers with retrieval to improve language models

Exploring language modelling at scale, including the use of Gopher and ethical considerations, for understanding and communication in artificial agents and humans.

A blogpost summarizing the papers and talks presented by Stanford AI Lab at NeurIPS 2021, including topics like improving compositionality of neural networks, reverse engineering recurrent neural networks, compositional transformers for scene generation, and more.

AI-assisted research in pure mathematics uncovers new insights in topology and representation theory, demonstrating the potential of machine learning in the field.

On the Expressivity of Markov Reward: A systematic study of the reward hypothesis and its limitations in conveying tasks to RL agents.