Jascha Sohl-dickstein
Member Of Technical Staff at Anthropic
Berkeley, California, United States
Summary
🤩
Rockstar🎓
Top SchoolJascha Sohl-dickstein is a senior research-minded software engineer with 15 years of experience building ML systems and user-facing software, currently a Member of Technical Staff at Anthropic in Berkeley. He progressed through research and engineering roles at Google—reaching Principal Research Scientist—after completing a PhD in Biophysics at UC Berkeley, blending rigorous academic training with production-grade engineering. His open-source contributions span front-end UX work for Khan Academy exercises to backend and DevOps fixes in BIG-bench and NLP transformation engineering in NL-Augmenter, showing comfort across the full stack of ML tooling. Notably, he has navigated tricky dependency and determinism issues in big collaborative benchmarks and implemented experimental UI interventions (A/B tests and metacognitive prompts) to study user engagement. He pairs a researcher's curiosity with practical software craftsmanship, often focusing on reproducibility, compatibility, and clear transformation APIs in shared repositories. Based in Berkeley, he brings deep domain knowledge in machine learning, experimental design, and scalable engineering to applied research teams.
15 years of coding experience
8 years of employment as a software developer
Doctor of Philosophy (PhD) Biophysics, Doctor of Philosophy (PhD) Biophysics at University of California, Berkeley
Github Skills (21)
dependency-manager10
dependency-management10
javascript10
back-end-development10
package-management10
python10
text-manipulation10
natural-language-processing10
nlp10
data-augmentation10
devops9
tensorflow9
regular-expression9
html9
test-automation8
Programming languages (6)
Github contributions (5)
google/BIG-bench
Jan 2021 - Oct 2022
Beyond the Imitation Game collaborative benchmark for measuring and extrapolating the capabilities of language models
Role in this project:
Back-end Developer & DevOps Engineer
Back-end Developer & DevOps EngineerContributions:283 reviews, 1350 commits, 408 PRs in 1 year 9 months
Contributions summary:Jascha primarily focused on improving the Python version compatibility and dependencies within the project. They modified the `setup.py` file to ensure compatibility with Python 3.8, removing support for newer versions of Python (3.9 and above) due to TensorFlow dependency issues. Additionally, they addressed minor issues related to determinism in tie-breaking for categorical accuracy. Furthermore, the user implemented the new option output_regex within the model generation functions.
language-model
Khan/khan-exercises
Aug 2012 - Apr 2014
A (deprecated) framework for building exercises to work with Khan Academy.
Role in this project:Front-end Developer
Front-end DeveloperContributions:16 commits in 1 year 8 months
Contributions summary:Jascha primarily contributed to the front-end of the project, focusing on enhancements and modifications to the user interface. Their work involved implementing a growth mindset A/B test, which included adding and removing motivational text within the interface. Additionally, they worked on a metacognitive experiment, adapting exercise files to include introspective and reflective questions within the problems, showing an understanding of UI/UX and user engagement. The user also performed some refactoring and merge operations.
khan-academy
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.