eigenslur means A theoretical concept according to which LLMs, by basing their output "calculating" the words, could find a pre-word (a token) that, when multiplied with another pre-word, always produces a slur.
eigenslur is pronounced /ˈaɪ.ɡənˌslɜː/.
Etymology
From eigen- + slur.
noun
- A theoretical concept according to which LLMs, by basing their output "calculating" the words, could find a pre-word (a token) that, when multiplied with another pre-word, always produces a slur.