engblogs

Exploring methods for verbalizing uncertainty in AI models

Exploring techniques for teaching models to communicate their uncertainty through natural language

A method called the Standardised Test Suite (STS) is developed for evaluating agents in multimodal interactions, allowing for control and speed in measuring progress.

Assessing existing evaluation metrics and introducing a novel approach called the Standardised Test Suite (STS) that uses real human interaction data.

Dynamic language understanding: adaptation to new knowledge in parametric and semi-parametric models

Kyrgyzstan to King’s Cross: the star baker cooking up code - A software engineer at DeepMind shares her journey from Central Asia to the tech industry and discusses her passion for learning, mentoring, and diversity.

This blogpost discusses the new features and updates in Chrome 103 beta, including early hints for navigation, origin trials, completed origin trials, and various other features.

The blogpost discusses the adaptation of parametric and semi-parametric models in dynamic language understanding to new knowledge.

A blogpost about the wide range of tasks a developer performs throughout different phases of a project, from designing solutions and deploying new features to communicating changes and fixing bugs.

The blogpost provides a list of accepted papers and talks from Stanford AI Lab at ACL 2022, covering topics such as pretraining language models, word order invariance, cosine similarity in embedding similarity, faithful summarization, spurious correlations in text generation evaluation, type-aware bi-encoders for entity retrieval, few-shot semantic parser for dialogues, representational harms, modular domain adaptation, and shared autonomy for robotic manipulation with language corrections.

Exploring innovative applications empowered by OpenAI Codex

How to build technology that benefits society and promotes responsible innovation in the field of artificial intelligence.