engblogs

Real-world challenges for AGI: tackling climate change and advancing the science of AGI

The blogpost discusses the real-world challenges that artificial general intelligence (AGI) can help solve, including climate change and other complex problems.

Exploring techniques to solve complex math word problems efficiently

Exploring the viability of factored cognition in language models through experiments using GPT-3 with decomposition to solve complex toy tasks and provide preliminary evidence.

The importance of education and hiring a developer educator at Jane Street.

Opening up a physics simulator for robotics: Advancing research everywhere with the acquisition of MuJoCo

Exploring the complexities of simulating physical contact for robotics research.

Selective classification can improve average accuracy but fail to improve or even hurt accuracy in certain subgroups of the data, highlighting potential failure modes of using selective classification in practice.

Discussing the powerful design choice of Apache Kafka and providing a sneak peek into its functionality.

Introducing RGB-Stacking as a benchmark for vision-based robotic manipulation and exploring how to enable robots to interact with a diverse range of objects.

Exploring the prevalent normalization methods for evaluating multiple choice tasks on autoregressive language models like GPT-3/Neo/J.

Exploring how to enable robots to better understand the interactions of objects with diverse geometries for more generalisable and useful robots.