Carnegie Mellon University's Language Technologies Institute (LTI) is seeking a Research Intern to develop Reasoning Reward Models (RRMs) that move beyond traditional scalar reward signals by generating natural language insights to explain human preferences. The goal is to create new methods for Large Language Model (LLM) post-training and alignment by bridging structured logical reasoning (e.g., Math) and open-ended human reasoning (e.g., Medical).
Requirements
- Programming proficiency in Python, with experience in deep learning.
- Knowledge and experience of LLM fundamentals.
- Academic or project-based familiarity with Large Language Model Concepts, including Supervised Fine-Tuning (SFT) and basic Reinforcement Learning (RL) concepts.
- Experience manipulating and processing datasets for NLP tasks (working with tokenizers, JSONL formats or Hugging Face datasets).
- Demonstrated ability to read technical research papers and implement algorithms or baselines from code repositories.
Responsibilities
- Train Reasoning Reward Models using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to generate natural-language insights for Math and Medical tasks.
- Implement insight-based post-training mechanisms, specifically 'Behavioral Priming' and 'Insight Exhibition Rewards,' to align models with interpretable reasoning criteria.
- Develop and implement software prototypes, algorithms, and data pipelines to support ongoing research projects.
- Execute experimental workflows and simulations, ensuring accurate data collection and logging.
- Analyze experimental results and metrics to identify trends, errors, or areas for optimization.
- Document technical processes, codebases, and research findings to ensure reproducibility and knowledge transfer.
- Maintain up-to-date knowledge of relevant tools, libraries, and best practices in software engineering and research.
Other
- Bachelor's Degree in Computer Science, AI, ML, Data Science or a related field.
- Flexibility and cultural sensitivity to work with a varied population of diverse audiences.
- Successful background check investigation may be required.
- Ability to work in Pittsburgh, PA.
- Position is a Staff – Fixed Term (Fixed Term) Full Time position with an Hourly pay basis.
