OpenAIOpenAI standardizes on PyTorch
Insights on OpenAI adopting PyTorch as their standard framework
SoundCloudSpeeding Up Builds with Dagger Reflect
Optimizing build times in Android apps by utilizing Dagger Reflect
InstagramFive things I learned about working on content quality at Instagram
The blogpost discusses five key learnings from working on content quality at Instagram, including the challenges of measuring quality, the importance of read-path models and sourcing-level filters, the need to track model performance, and the use of calibration and percentile frameworks.
OpenAIScaling laws for neural language models
Exploring the scalability of neural language models through scaling laws
SoundCloudThe Power of Types for Errors
A Kotlin-centric exploration of replacing loosely coupled error parameters with a typed error hierarchy (sealed classes) to boost compile-time safety, readability, and domain-specific error handling.
SoundCloudThe Power of Types for Errors
A Kotlin-focused guide on using custom types and sealed class hierarchies to model errors, replacing mixed primitive params and enums to improve safety and readability.
SoundCloudThe Power of Types for Errors
Exploring the benefits of using types to minimize the usage of primitives in code
Jane StreetUsing Python and OCaml in the same Jupyter notebook
A guide on how to use Python and OCaml together in a Jupyter notebook
OpenAIDota 2 with large scale deep reinforcement learning
Exploring the application of large scale deep reinforcement learning in Dota 2
InstagramInstagram Data Saver Mode
The blogpost discusses the new Data Saver Mode feature on Instagram for Android that reduces mobile data consumption.
Jane StreetDeep-Learning the Hardest Go Problem in the World
Updates and a New Run
OpenAIDeep double descent
Exploring the phenomenon of deep double descent in machine learning