Questions to LLMs panel @ RL+LLM workshop @ AAAI 2024
saved
Ideas
Pros and cons
Votes
Hallucination is good to some degree to help LLMs extrapolate and generalize. How RL can help with mitigating false hallucinations while encouraging correct hallucinations?
1
Vote
Mohamad
What are challenges and opportunities for NLP researchers with limited resources, e.g., most university profs and PhD students, in the LLMs era? What may be great research topics for them?
1
Vote
Maryam Hashemi
How can reinforcement learning techniques be leveraged to enhance interpretability and explainability in language models, ensuring transparency and trustworthiness in their decision-making processes?
by Maryam Hashemi
0
Vote
What are the main obstacles to improve DPO results vs RLHF? Is overfitting the only issue with DPO (which can be solved by a huge labeled dataset), or does learning the reward and policy simultaneously create other issues?
by Salarrahili
0
Vote
The RLHF framework has challenges with costly data annotators and sparse rewards. How can we leverage offline data to create a denser reward function?
by Elahe Aghapour
0
Vote
How can integrating RL facilitate the development of personalized LLMs for individual user preferences? Furthermore, what implications does it raise regarding privacy concerns?
by Mojtaba
0
Vote
What are the main challenges for LLMs to have solid planning capabilities? How can reinforcement learning help?
0
Vote
How can LLMs weigh the value of seeking more information from the user vs. responding the query with existing information? Thoughts on fusing exploration techniques from reinforcement learning?
0
Vote
How can LLMs personalize themselves through the user feedback to provide a more refined experience for the user?
0
Vote
How do you like approaches like self-generation, self-evaluation, and self-improvements (based on a fixed LLM)? Is it information perpetual motion machine?
0
Vote
Will GPT lead to AGI? Any suggestions for potential alternatives?
