[shantara]

What terrifies me is the idea of someone building a GPT-based bot specifically targeting Wikipedia. If one would train a model on the existing wiki dataset, it could generate and submit plausibly looking but completely factually false edits and articles with plausibly looking citations. Given the history of long-lasting hoaxes, it shouldn’t be hard to achieve enough throughput to completely overwhelm the capacity of human editors to do any fact checking.