Weijia Shi

Weijia Shi works in pretraining engineering at Anthropic. Her research develops augmented and modular language-model architectures and training methods, including approaches that use external tools and specialized models to improve factuality, knowledge access, and control. She earned a Ph.D. from the University of Washington, where her work included the thesis *Breaking the Language Model Monolith*.

Career

Cosigned byLoading...

CosignedSign in to view

Elsewhere

Anthropic

Sep 2026 — Present

Pretraining

University of Washington

Ph.D.

Products & media

iclr.cc

Knowledge Card: Filling LLMs’ Knowledge Gaps with Plug-in Specialized Language Models

A modular approach to adding domain-specific knowledge to general-purpose language models, with methods for selecting and integrating relevant information.

Writing

nlp.stanford.edu

Beyond Monolithic Language Models

A seminar talk about using external data and tools, and modular model designs, to address reliability, updating, privacy, and copyright challenges.

Media appearance

digital.lib.washington.edu

Breaking the Language Model Monolith

Her University of Washington doctoral thesis examines external and internal modularity for language models, including tools for factuality and reasoning.

Writing

export.arxiv.org

REPLUG: Retrieval-Augmented Black-Box Language Models

A research paper describing a retrieval framework that augments frozen language models with retrieved documents.

Writing

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Weijia Shi

Anthropic

University of Washington

Products & media

Knowledge Card: Filling LLMs’ Knowledge Gaps with Plug-in Specialized Language Models

Beyond Monolithic Language Models

Breaking the Language Model Monolith

REPLUG: Retrieval-Augmented Black-Box Language Models

Curated Lists

List Inclusions