engblogs

This role has been filled

Enhancing language model performance through training on a carefully curated dataset

An update on DeepMind's racial justice efforts and intention to combat racism and advance racial equity.

Exploring the benefits of creating and open sourcing a large language model for AI safety

A practical blueprint for safely upgrading a central library (JVMKit) across SoundCloud’s polyglot stack by using a reverse dependency graph, Golden Paths, and automated CI pipelines to verify each dependent service with batch upgrades and targeted tests.

A practical approach to automatically verifying JVMKit changes against SoundCloud’s ecosystem by tracing reverse dependencies through a graph, enforcing Golden Paths, and running CI-driven pipelines to ensure safe, disruption-free releases.

Exploring the process of reverse dependency verification and its importance in maintaining a tech stack amidst changing company dynamics

Examining the impact of removing activation functions in GPT-like autoregressive language models.

Tuning the performance of GPT-Neo on downstream tasks using an evaluation harness.

Evaluating the impact of different fewshot prompts on the performance of GPT-3.

Utilizing eval harness to determine the sizes of OpenAI API models based on their performance.

Exploring innovative final projects from OpenAI Scholars 2021 cohort.