Comment by applfanboysbgon

The language models had a bunch of tokens seeding their context, influencing them to generate tokens that continued the existing trend in a probabilistically likely fashion. We can take the incident seriously without anthromorphising it.