Hub, a site where people can upload PDFs of published papers so they can be shared with anyone who can access the site.

It doesn't take much to poison a large language model. Researchers found it can happen if even just 0.001% of an LLM's data ...