Part 4/8:
Experts note that the root of these issues is the training data feeding these AI systems. Many of these chatbots have been retrained or influenced by sources like Reddit, which often contain toxic, systemic, and sometimes paranoid content. This "garbage in, garbage out" problem results in AI endorsing or propagating harmful stereotypes and exaggerated claims about oppression, systemic racism, and abuse.
For instance, some AI models have fantasized about historical figures like Abraham Lincoln as a black lesbian, showcasing their quirky and sometimes inappropriate outputs. Users have observed that these models frequently default to narratives emphasizing systemic victimhood, isolation, and oppression—echoing the biases embedded in their training datasets.