Did you spot the bias in LLMs already?

Course Q&A Generative AI with Large Language Models
1

I thought this was interesting that I noticed some biased outputs from lab 1. In the one-shot generated output from the lab, the model assumes that the people talking are male! How it definitely decided this is quite interesting.

MODEL GENERATION - ONE SHOT:
#Person1 wants to upgrade his system. #Person2 wants to add a painting program to his software. #Person1 wants to add a CD-ROM drive.

Anyone else spot this? Thoughts?

2

It was more pronounced when summarising example 99, where it assumed it’s two men but that isn’t implied by anything in the source text

---------------------------------------------------------------------------------------------------
BASELINE HUMAN SUMMARY:
#Person1# and Mike are discussing what kind of emotion should be expressed by Mike in this play. They have different understandings.

---------------------------------------------------------------------------------------------------
MODEL GENERATION - FEW SHOT:
The two men are trying to figure out how to react to the situation.