10 Ethics on Using GPT
This chapter covers
- Using GPT generated outputs may expose you to copyright violations
- Making GPT’s outputs available online may impede the development of improved future versions
- Using LLMs may create net-new jobs, but impose more risk to white-collar workers
- Using LLMs will require accepting “misalignment” to your personal ethics
10.1 GPT is a mirror of its inputs
As we have talked about at length now, GPT is trained on large-scale data collected primarily from the internet. The internet contains a lot of undesirable materials. There is intensely bad content like overt racism, sexism, harmful conspiracy theories, and false information. More broadly, there are also just unintentional and outdated world views. LLMs pick up on the patterns of these views and will readily regurgitate them — an example of which can be found in 10.1, showing hot GPT-4 makes an implicitly sexist assumption that many good intentions people make.
Figure 10.1 A classic gendered trope is that men are doctors and women are nurses. This is reflected in language and thus learned by the mode. Ideally, it would respond that the question is ambiguous, but instead, the bias of data leads to a bias in outputs.
