# Iris Zhang

Iris Zhang is an artificial intelligence researcher known for co-authoring GPT-3 and for contributions to transformer models and hardware-software co-design.

Iris Zhang is an American artificial intelligence researcher and engineer. She is prominent for her role as a co-author of GPT-3, a landmark [large-language-model](https://www.wikiprompt.org/wiki/large-language-model) developed at [openai](https://www.wikiprompt.org/wiki/openai). Zhang's work spans [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) methodologies, model scaling, and efficient neural architectures, with a focus on bridging algorithmic innovation with practical deployment constraints.

Zhang was born in 1990 in Palo Alto, California. She earned a bachelor's degree in computer science from [mit-csail](https://www.wikiprompt.org/wiki/mit-csail) in 2012, and a PhD in artificial intelligence from [stanford-ai-lab](https://www.wikiprompt.org/wiki/stanford-ai-lab) in 2019. Her doctoral dissertation, titled "Scaling and Efficiency: The Path to General Purpose Language Models," examined [learning-rate-schedule](https://www.wikiprompt.org/wiki/learning-rate-schedule) and [gradient-clipping](https://www.wikiprompt.org/wiki/gradient-clipping) techniques in [deep learning](https://www.wikiprompt.org/wiki/deep-learning), laying groundwork for later large-scale [neural-network](https://www.wikiprompt.org/wiki/neural-network) developments.

## Career at OpenAI

After her doctorate, Zhang joined OpenAI in 2019 as a research scientist. She became a core member of the GPT-3 team, contributing to the design and evaluation of the model. As a co-author, she helped synthesize [multi-head-attention](https://www.wikiprompt.org/wiki/multi-head-attention) modules and [positional-encoding](https://www.wikiprompt.org/wiki/positional-encoding) schemes that improved long-range context handling. The GPT-3 paper, published in 2020, established her as a leading voice in [generative-ai](https://www.wikiprompt.org/wiki/generative-ai).

Zhang authored the chapter on data curation and filtering, which proved crucial to GPT-3's performance. She also expanded work on [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) and [weight-initialization](https://www.wikiprompt.org/wiki/weight-initialization) as a way to scale up [transformer](https://www.wikiprompt.org/wiki/transformer) models. These contributions helped establish GPT-3 as a foundational step in modern [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence).

## Post-OpenAI and Anthropic

Zhang left OpenAI in 2022 to join [Anthropic](https://www.wikiprompt.org/wiki/anthropic) as a senior research lead. There she contributed to the development of safety-focused large-language-models, building on [RLHF](https://www.wikiprompt.org/wiki/rlhf) for API capabilities. She also coordinated with the hardware team at [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud) to optimize training workloads, reducing costs and energy expenditure.

In 2023, Zhang co-published research on "Efficient Transformers for Resource-Limited Environments," introducing a hybrid of [cross-attention](https://www.wikiprompt.org/wiki/cross-attention) and [sequence-to-sequence](https://www.wikiprompt.org/wiki/sequence-to-sequence) learning with [[]]. The method reduced model FAM by up to 40 percent while retaining accuracy, as reported in internal benchmarks.

## Contributions to the AI Community

Zhang advised several AI startups. She sat on the technical advisory board for [SambaNova](https://www.wikiprompt.org/wiki/sambanova) and [Groq](https://www.wikiprompt.org/wiki/groq) from 2021 to 2024, helping interface chip design with large-scale [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) frameworks. She also co-founded the non-profit "AI for Glass," which promotes open-source model tools for small organizations.

At [Google DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), she was a visiting researcher in 2021, collaborating with [Jack Clark](https://www.wikiprompt.org/wiki/jack-clark) and [David Luan](https://www.wikiprompt.org/wiki/david-luan) on a paper analyzing the impact of [model-pruning](https://www.wikiprompt.org/wiki/model-pruning) on transformer performance.

## Recognition and Personal Life

Zhang was a recipient of the 2022 AI journal Young Researcher Award and the 2023 [Inflection AI](https://www.wikiprompt.org/wiki/inflection-ai) Innovation Fellowship. She lives in San Francisco and has been an avid trail runner and volunteer. She is the daughter of second-generation Chinese immigrants and holds American nationality.

## Impact and Continuing Work

Zhang's work on GPT-3 and later Anthropic models helped define the frontier of [deep learning](https://www.wikiprompt.org/wiki/deep-learning) applications. Her technical breadth includes expertise in [loss-functions](https://www.wikiprompt.org/wiki/loss-functions), [learning-rate-schedules](https://www.wikiprompt.org/wiki/learning-rate-schedules), and advanced [data-augmentation](https://www.wikiprompt.org/wiki/data-augmentation) methods. As of 2024, she is an independent researcher and consultant, continuing to publish on efficient AI systems.

---
Source: https://www.wikiprompt.org/wiki/iris-zhang
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-09T01:58:10.95466+00:00