HAI Postdoctoral Fellowship with Professor Chris Potts, Department of Linguistics

Foundation Models (FMs) are enabling researchers to build AI systems at higher levels of abstraction and with lower data requirements than ever before. However, these systems are being built around hand-crafted, prompt templates. This is akin to setting the weights of a classifier by hand rather than learning them from data.

Toward a more systematic approach, we have introduced DSPy, a programming model that abstracts FM pipelines as imperative computation graphs. DSPy modules are parameterized, meaning they can learn how to apply compositions of prompting, fine-tuning, augmentation, and reasoning techniques, and the DSPy compiler will optimize any DSPy pipeline. Even simple DSPy pipelines, once compiled, routinely outperform pipelines with hand-created prompts and allow us to develop performant systems using relatively small LMs.

We see at least four main areas in which the postdoctoral fellow might contribute: (1) contributing to the DSPy library, (2) developing new FMs, (3) creating new approaches to optimizing DSPy programs, and (4) explaining why specific DSPy programs are effective using interpretability techniques we have developed. The postdoctoral fellow will be free to chose which of these areas to focus on, and we are open to new areas we might be overlooking right now.

Mentorship structure

The postdoc will primarily be supervised by Potts, with informal supervision from Matei Zaharia (Berkeley) and Omar Khattab (Stanford). The postdoc will also become a member of the Stanford NLP Group.

Stanford HAI is also committed to creating a diverse community of scholars who are engaged in contributing to the understanding and advancement of Human-Centered AI. Postdoctoral fellows will have the opportunity to engage with one another and with the broader Stanford HAI research community. They are also expected to participate in professional development, cohort-building, and other programmatic activities organized by HAI.

Qualifications

  • Research experience with foundation models, neural information retrieval, in-context learning, and optionally interpretability techniques and explainable AI
  • Interest in open-source software development
  • A record of scientific collaboration and mentorship

Evaluation criteria

  • Research excellence
  • Alignment of research interests with project goals
  • Track record of mentoring more junior scholars

Timeline

  • Application Deadline: March 20, 2024 (Applicants advancing in the review process may be asked to submit additional materials, including letters of recommendation, and may be invited to interview.) 
  • Selections to be made by mid-April, 2024
  • This is a 1-year appointment starting Fall 2024

Apply

For full consideration, send a complete application in a single PDF to HAI-Fellowships@stanford.edu  with the subject line: “HAI Postdoc Fellowship - Potts”

Complete applications will include:

  • Cover letter explaining your interest in becoming a postdoc with the respective research lab/center/faculty through the HAI Postdoctoral Fellowship Program (500 words max).
  • Curriculum Vitae (CV)
  • Short answer responses (300 max each):
    • Description of your dissertation research and broader research agenda
    • How do you see your research training and expertise contributing toward the understanding and advancement of Human-Centered AI? HAI thinks about "Human-Centered AI" through three focus areas.
  • One representative writing sample (published or unpublished)

The expected base pay for this position is $83,600/yr. The pay offered to the selected candidate will be determined based on factors including (but not limited to) the qualifications of the selected candidate, budget availability, and internal equity.

Stanford is an equal opportunity employer and all qualified applicants will receive consideration without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, or any other characteristic protected by law.