Reinforcement Learning for Language Models

By G. Qian, Wanqiao Xu, P. Dupenloup, Jesse Mu. 2023

Subfields: Large Language Models, Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Large Language Models

1. Sergey Levine
2. Charles Burton Snell
3. Dawn Song
4. Eric Wallace
5. Pieter Abbeel

In Reinforcement Learning

1. Sergey Levine
2. Pieter Abbeel
3. Chelsea Finn
4. Roberta Raileanu
5. Abhishek Gupta
6. Vikash Kumar
7. Shixiang Shane Gu
8. Edward Grefenstette

Citations in the corpus

• The False Promise of Imitating Proprietary LLMs (2023, background): “…that during finetuning time, you may even want to avoid introducing new knowledge (i.e., do not imitate better models), as you will otherwise be training the model to guess or hallucinate its answers, rather than actually doing the task as intended (Schulman, 2023; Gao, 2021; Goldberg, 2023).”
• Understanding the Effects of RLHF on LLM Generalisation and Diversity (2023, background): “Also, while there are some hypotheses about why RLHF generalises better than SFT (Goldberg, 2023), it is important to experimentally validate these in order to build our understanding of how these methods work and when they should be used.”
• Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, unknown): “However, when trained on instructions for which LLMs have not acquired prerequisite knowledge from the pretraining phase, this is actually a misalignment process that encourages LLMs to hallucinate (Goldberg, 2023; Schulman, 2023).”