Jon Borchardt

Also known as Jonathan Borchardt

Jon Borchardt is an AI builder whose work spans scientific search, machine ethics, open vision-language models and tools for tracing language-model data. He earned a B.S. from the University of Washington in 2001, then coauthored SciSight, Delphi and ACCoRD. His later work includes the Molmo family, olmOCR and OLMoTrace, whose paper won the ACL 2025 Best Demo award. As of October 2026, Jonathan Borchardt is a member of Ai2’s team.

Now at: Allen Institute for AI (Ai2)

Timeline

2001

University of Washington degree

He earns a B.S. from the University of Washington in 2001.
May 2020

SciSight for COVID-19 research

In May 2020, he coauthors SciSight with Tom Hope, Jason Portenoy, Kishore Vasan, Eric Horvitz, Daniel S. Weld, Marti A. Hearst and Jevin West. The system explores biomedical facets such as genes, drugs and diseases, then combines textual and network information to find groups of researchers and their ties. The paper is accepted to EMNLP 2020.
October 2021–January 2025

The Delphi experiment

In October 2021, he coauthors “Can Machines Learn Morality? The Delphi Experiment” with Liwei Jiang, Yejin Choi and colleagues. Delphi is a deep neural network framework trained to reason about descriptive ethical judgments. Nature Machine Intelligence publishes the study in January 2025 with Borchardt as an Allen Institute for Artificial Intelligence author in Seattle; Delphi is an open-source system trained to predict US participants’ moral judgements.
May 2022

The ACCoRD corpus

In May 2022, he coauthors ACCoRD with Sonia K. Murthy, Kyle Lo, Tom Hope, Doug Downey and colleagues. The end-to-end system generates diverse descriptions of scientific concepts, and the team releases a corpus of 1,787 hand-authored concept descriptions.
2024–2025

Molmo and PixMo

In September 2024, he coauthors Molmo and PixMo with Matt Deitke, Ross Girshick, Ali Farhadi, Aniruddha Kembhavi and colleagues. Molmo is a family of open vision-language models, and PixMo is a collection of new datasets gathered without external VLMs; its best 72B model outperforms Claude 3.5 Sonnet and Gemini 1.5 Pro. CVPR publishes the paper in 2025.
February 2025

olmOCR for PDF text

In February 2025, he coauthors olmOCR with Jake Poznanski, Kyle Lo, Luca Soldaini and colleagues. The open-source toolkit runs a fine-tuned 7B vision-language model to convert PDFs into clean, linearized plain text for language-model training. The AllenNLP team, backed by Ai2, maintains the toolkit under Apache 2.0.
2025

OLMoTrace design and award

In 2025, he is one of four design contributors to OLMoTrace, a system that traces language-model outputs back to its full, multi-trillion-token training data in real time. The paper wins the ACL 2025 Best Demo award.
As of October 2026

Team member at Ai2

As of October 2026, Jonathan Borchardt is a member of Ai2’s team.