Skip to main content

Towards A Better Understanding of Language Modeling and Reasoning

Machine Learning Lunch Meeting: Junjie Hu, Tuesday May 2, 12:15pm CS 1240

Event Details

Date
Tuesday, May 2, 2023
Time
12:15 p.m.
Location
Description

You are cordially invited to the weekly CS Machine Learning Lunch Meetings. This is a chance to get to know machine learning professors, and talk to your fellow researchers.  Our next meeting will be on Tuesday May 2 12:12-1:30pm in CS 1240. Professor Junjie Hu will reason about GPT-4, see abstract below.

If you would like to be informed of future CS Machine Learning Lunch Meetings, please sign up our mailing list at https://lists.cs.wisc.edu/mailman/listinfo/mllm -- please use your cs or wisc email.  After you enter your email, the system will send you an email for confirmation.  Only after you respond to that email will you be on the mailing list.

Abstract: Over the past few years, the phenomenal success of NLP systems, such as ChatGPT and GPT-4, has been driven by the pre-training of large language models (LLMs) on massive raw texts. While LLMs have been booming, concerns about their reliability, especially in complex, unseen scenarios, have also arisen. In this talk, I will discuss several research questions: (1) How do pre-trained LLMs learn transferable language features? (2) How can we prompt LLMs to generate multi-step reasoning paths for complex language questions? I will also present our lab's ongoing work on LLMs' multi-step reasoning capability and discuss our initial findings. Finally, I will highlight several open research questions.

Cost
Free

Tags