Sections

Project List

Devise hybrid systems combining crowdsourcing and LLMs to improve misinformation detection using HITL techniques and Think-Reason-Act prompting.
Analyze misinformation with Graph Neural Networks (GNNs) and develop models for tracking misinformation using crowdsourcing data.
Apply RLHF to improve misinformation detection models, integrating human feedback and exploring ethical considerations and biases.
Develop custom autoregressive models to generate explanations for LLM-based misinformation systems, enhancing transparency and trust.
Create retrieval-based systems for misinformation detection using Retrieval Augmented Generation (RAG), balancing recall and precision.
Fine-tune LLMs for query generation, optimizing custom objective functions and producing diverse, high-quality queries.
Design specialized prompts for industrial applications using LLMs to enhance efficiency and address real-world challenges.
Study and mitigate biases in LLMs by developing bias-free models through alignment and evaluating their impact.
Probe LLMs for fact-checking properties and domain-specific understanding using custom probing frameworks.
Build domain-specific LLMs with tailored pretraining and alignment strategies, focusing on medical and misinformation domains.