Trustworthy AI

Trustworthy Code Generation

Measuring and controlling what LLMs generate

Active Research

Overview

Incidental cues in a prompt can change which algorithm a model chooses by up to 100 percentage points, among solutions that pass the same tests. Standard benchmarks, which check only outcomes, cannot detect this. We study how context steers generated code, who decides what counts as correct when agents repair programs, and how to build LLM pipelines for code and SQL that can be checked.

Related Publications

The Invisible Lottery (ICML 2026)
Whose Oracle Is It? (APR 2026)
SQL-of-Thought (DL4C @ NeurIPS 2025)
LoRACode (DL4C @ ICLR 2025)