Predicting vs. acting: A trade-off between world modeling & agent modeling
RLHF-aligned LMs have shown unprecedented ability on both benchmarks and long-form
text generation, yet they struggle with one foundational task: next-token prediction. As RLHF …
text generation, yet they struggle with one foundational task: next-token prediction. As RLHF …
Dual Active Learning for Reinforcement Learning from Human Feedback
P Liu, C Shi, WW Sun - arXiv preprint arXiv:2410.02504, 2024 - arxiv.org
Aligning large language models (LLMs) with human preferences is critical to recent
advances in generative artificial intelligence. Reinforcement learning from human feedback …
advances in generative artificial intelligence. Reinforcement learning from human feedback …
Entropic Distribution Matching in Supervised Fine-tuning of LLMs: Less Overfitting and Better Diversity
Large language models rely on Supervised Fine-Tuning (SFT) to specialize in downstream
tasks. Cross Entropy (CE) loss is the de facto choice in SFT, but it often leads to overfitting …
tasks. Cross Entropy (CE) loss is the de facto choice in SFT, but it often leads to overfitting …
Fine-Tuning Linear Layers Only Is a Simple yet Effective Way for Task Arithmetic
Task arithmetic has recently emerged as a cost-effective and scalable approach to edit pre-
trained models directly in weight space, by adding the fine-tuned weights of different tasks …
trained models directly in weight space, by adding the fine-tuned weights of different tasks …
State of Title IX: A Knowledge Base for Title IX Documentation
P Sharma - 2024 - search.proquest.com
Abstract Knowledge bases play a vital role in the modern world, offering a systematic and
structured approach to integrate various entities, concepts, rules, and relationships …
structured approach to integrate various entities, concepts, rules, and relationships …