engblogs

Challenges in Detoxifying Language Models: This blogpost discusses the challenges in mitigating toxic language generation in language models and highlights the unintended consequences of detoxification.

Exploring methods to mitigate toxicity in language models and evaluating their effectiveness and limitations using classifier-based automatic toxicity evaluation.

Exploring model accuracy in replicating human misconceptions through TruthfulQA

Announcement of Helen Toner's inclusion in OpenAI's board of directors.

An alternative approach to digital identity that replaces user profiles with trusted digital relationships.

Restructuring standard libraries to eliminate difference between Core_kernel and Core at Jane Street with open source release anticipated by end of year

Examining the evolution of service architecture at SoundCloud with a focus on value-added services.

Comparing Rotary Position Embeddings with GPT-style learned position embeddings in downstream evaluations.

Exploring the capabilities of OpenAI Codex in automation and natural language processing

Internship season at Jane Street in 2021, including in-person interactions and insights into the company's operations.

Building a general-purpose architecture, Perceiver IO, that can handle diverse inputs and produce a wide variety of outputs, enabling the processing of various types of data.

Exploring a general and versatile architecture that can handle different types of data to solve complex problems in AI systems.