Jascha Sohl-Dickstein

Career

1. 
   Anthropic

   • Member of Technical StaffFeb 2024 – present
2. 
   Google

   • Senior → Staff → Senior Staff → Principal Research ScientistAug 2015 – Jan 2024
3. 
   Khan Academy

   • Academic Resident2012–2015
4. 
   Stanford University

   • Visiting Scholar2012–2015
5. 
   NASA Jet Propulsion Laboratory

   • Research associate, Mars Exploration Rovers2001–2005

Education

• Doctor of Philosophy (PhD), Biophysics, University of California, Berkeley, 2005-2012

Lesser-known facts

• PhD in biophysics from UC Berkeley.
• Was an academic resident at Khan Academy.
• Co-authored the 2004 Science papers on the Pancam cameras of the Spirit and Opportunity Mars rovers while at NASA JPL.
• Earned a second Cornell BA, as a College Scholar, in quantum computing alongside his physics degree.
• Co-wrote "Deep knowledge tracing" (NeurIPS 2015), which grew out of his education work at Khan Academy.
• Wrote the original design doc for Google's BIG-bench benchmark and was its top GitHub contributor.
• Proposed "the hot mess theory of AI misalignment" in a 2023 blog post, before joining Anthropic.
• Showed in a viral 2024 post that the boundary between trainable and untrainable hyperparameters is fractal.

Known for

• Deep Unsupervised Learning using Nonequilibrium Thermodynamics (2015)
• Score-Based Generative Modeling through Stochastic Differential Equations
• Density estimation using Real NVP
• Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models (2022)
• Position: Levels of AGI for Operationalizing Progress on the Path to AGI (2024)

Interviews and talks

• Talk: Advice for a young investigator in the first and last days of the Anthropocene (2025)x.com