Description
MIT Sloan School of Management | Faculty supervisor: Eric So | Funded by the MIT Generative AI Impact Consortium (MGAIC)
Purpose of the appointment
The Postdoctoral Associate will join a research program on how AI systems can be built, evaluated, and deployed in large-scale investment decisions. These decisions require synthesizing large volumes of heterogeneous information (financial statements, diligence reports, management materials, market data) into a judgment about value and risk. The program asks when AI systems add real decision value, how to measure what information they rely on, and where they fail. Initial work is funded by the MIT Generative AI Impact Consortium (MGAIC) in partnership with a leading global investment firm. It uses historical private-market investment records paired with realized outcomes to test whether models identify the information that actually drives investment decisions. The postdoc will help shape where the agenda goes next.
The appointment ties directly to Professor So's applied AI research on evaluating and probing large language models in high-stakes decision settings. That work includes designing benchmarks and evaluation protocols, running controlled experiments on frontier models through their APIs, measuring model reliance on specific inputs through ablation and perturbation, and detecting memorization and data leakage. The postdoc will lead the technical side of this work: building evaluation pipelines and experiments, producing research papers, and acting as the day-to-day technical link between MIT and the industry partner's research team.
Responsibilities
• Build benchmark instances from confidential diligence records: point-in-time controls, document and claim-level parsing, and outcome definitions fixed before model evaluation.
• Implement the selector-forecaster evaluation pipeline. A selector model picks a fixed-size evidence subset, and a separate forecaster model is scored with proper scoring rules (Brier and log scores).
• Run sufficiency and comprehensiveness tests against random-subset and no-document baselines, and use those baselines to diagnose memorization and recognition leakage.
• Develop the agreement-with-deal-team track: claim-level matching against investment committee records and document-type baselines.
• Probe and compare frontier models systematically through their APIs: design prompts, run ablation and perturbation experiments, and manage cost, rate limits, and reproducibility.
• Co-author academic papers and write practitioner-facing summaries. Present results at MGAIC and partner meetings, and at academic conferences.
• Help supervise research assistants (PhD, MS, or UROP) on annotation and data work.
Qualifications
Required
• PhD in computer science, economics, or finance, completed by the start date.
• For economics and finance candidates: a clear orientation toward machine learning and applied AI research, shown by papers, code, or projects.
• Strong programming skills in Python and experience building reproducible empirical or ML pipelines.
• Hands-on experience with large language models, including running experiments through model APIs.
• Ability to work independently and handle confidential data under a data-use agreement.
Strongly preferred
• Experience with model probing through APIs: systematic behavioral testing of LLMs, evaluation harnesses, log-probability or calibration analysis, and input ablation or perturbation.
• Background in benchmark or evaluation design, retrieval and attribution, rationale evaluation, or forecasting and scoring rules.
• Familiarity with financial statements, valuation, or investment decision-making (prior industry experience is welcome but not required).
• A record of research publications or strong working papers.
Appointment terms
• Title: Postdoctoral Associate, MIT Sloan School of Management.
• Start date: Flexible; ideally as soon as possible.
• Term: 12 months, renewable subject to funding and performance.
• Location: On campus at MIT Sloan, Cambridge, MA.
• Salary: Competitive, at or above the MIT postdoctoral minimum, with full MIT benefits. The salary range for this position is $73,308-$90,000.
Application Instructions
Send a CV, a one-page cover letter describing relevant research and experience working with LLMs, one or two writing samples or papers (links to code repositories welcome), and the names of two or three references to Eric So ([redacted]) with the subject line "MGAIC Postdoc Application." Applications are reviewed on a rolling basis until the position is filled.
