engblogs

summaries of the latest blog articles from your favorite tech companies.
Google DeepMindGoogle DeepMind

Alchemy: A structured task distribution for meta-reinforcement learning

Introducing Alchemy, a benchmark environment and analysis tools for meta-reinforcement learning.

2/8/2021
OpenAIOpenAI

Understanding the capabilities, limitations, and societal impact of large language models

Exploring the impact of large language models on society and their capabilities and limitations

2/4/2021
SoundCloudSoundCloud

Tests Under the Magnifying Lens

Explains SoundCloud's shift from traditional test reporting to a data-driven approach where test results feed build pipelines, address flakiness and pipeline delays, and fuel Grafana-powered dashboards for cross-system visibility.

2/2/2021
SoundCloudSoundCloud

Tests Under the Magnifying Lens

A data-driven tour of SoundCloud's testing strategy, from the test pyramid and cross-repo data aggregation to rendering dashboards that reveal flakiness and speed up build pipelines.

2/2/2021
SoundCloudSoundCloud

Tests Under the Magnifying Lens

The importance of testing in engineering practices at SoundCloud and building well-balanced test pyramids within code repositories.

2/2/2021
Google DeepMindGoogle DeepMind

Data, Architecture, or Losses: What Contributes Most to Multimodal Transformer Success?

Examining the importance of multimodal attention in the success of multimodal transformers for pretraining.

2/2/2021
OpenAIOpenAI

Scaling Kubernetes to 7,500 nodes

Efficient strategies for scaling Kubernetes deployments to 7,500 nodes

1/25/2021
OpenAIOpenAI

CLIP: Connecting text and images

Exploring the integration of text and images through CLIP technology

1/5/2021
OpenAIOpenAI

DALL·E: Creating images from text

Exploring how DALL·E generates images based on textual descriptions

1/5/2021
OpenAIOpenAI

Organizational update from OpenAI

Exploring the latest organizational developments at OpenAI

12/29/2020
Google DeepMindGoogle DeepMind

MuZero: Mastering Go, chess, shogi and Atari without rules

MuZero is a general-purpose algorithm that can master Go, chess, shogi, and Atari without being given the rules, using its ability to plan winning strategies in unknown environments.

12/23/2020
Google DeepMindGoogle DeepMind

Imitating Interactive Intelligence

Creating a simulated environment (Playroom) for virtual robots to engage in interactive intelligence, studying joint intentionality, cooperation, and communication of private knowledge.

12/11/2020