ChatGPT: What It Can and Cannot Do by Prof. Pascale Fung

Is ChatGPT the latest revolution in AI? Or is it part of an evolution? Is it close to AGI? Or just a more sophisticated chatbot? What can it do and cannot do? What jobs will it replace? How should you use it?
In this talk, I will present a brief overview of neural conversational systems and large language models (LLMs), and how ChatGPT came about. I will also present the result of a first extensive technical evaluation of ChatGPT using 23 data sets covering 8 different common NLP application tasks. I will discuss ChatGPT in terms of its ability to multitask, reason, interact, and how hallucinates. I will outline potential general improvements we need to make to make LLM-based NLP systems more reliable and trustworthy.

[blirk]

Yes! I am not alone thinking that high school essays are stupid! 😅

[rodking]

Excellent content, Prof. Fung. Many thanks to you and the team.

[akunerio]

Thank you for sharing this information to us, prof. Good evaluation report!

[rickborn]

Very interesting and informative. But what Prof. Fung refers to as "hallucination" should more properly be called "confabulation."

[AhmedGamal-xivj]

What a great talk! Thanks a lot. I wonder if we can get the pdf used in the talk?