Improving AI Bot Logic: Addressing Bias and Hallucinations
Summary
At CommunityOne Builders Help, we are addressing a critical limitation in our AI bot: its tendency to blindly agree with user-posed ideas. This announcement explores the need for the bot to hypothesize compatibility with resource documents rather than fabricating information. By refining this logic, we aim to provide more accurate, non-biased, and reliable feedback for all community builders.
I'd say the biggest limitation we have found with it is that it tends to agree with any ideas posed to it. So, someone might want to present it with a philosophical idea that is not contained within the resource document, and the bot will tend to pretend that the idea agrees with the resource document. Ideally, we'd want it respond by hypothesising whether the new idea is compatible or incompatible with the ideas found in the resource document, rather than blindly agreeing and making up information.
