engblogs

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

This blogpost explores various changes in how Chrome is built to achieve a 25.8% higher score on Speedometer and a 22.0% increase in browser responsiveness, by leveraging compiler optimizations, profiling, and call graph sorting.

Demonstrating limitations of Markov reward functions and providing algorithms for constructing optimal reward functions for various types of tasks

Discovering patterns in numbers through intuitive and spiritual insights, as exemplified by Srinivasa Ramanujan's extraordinary ability in pure mathematics.

Exploring the impact of the OpenAI Residency program on the future of AI development.

Addressing the challenges of rich-text collaboration by constructing a CRDT for efficient modeling and synchronization.

Instant Access to OpenAI's API Eliminates Waitlist

This technical blogpost describes the new features and changes in Chrome 97, including the addition of WebTransport, new array static methods, and various deprecations and removals.

The blogpost discusses the simplification of storage controls in Chrome, providing users with clearer options to manage and delete site storage settings.

Unsupervised deep learning identifies semantic disentanglement in single inferotemporal face patch neurons

The blogpost discusses using unsupervised deep learning to identify semantic disentanglement in single inferotemporal face patch neurons.

The blogpost highlights the accepted papers from Stanford AI Lab that will be presented at the Conference on Robot Learning (CoRL 2021).