79% of AI failures in human interactions are invisible to users and form eight stable archetypes that persist even as models improve.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Invisible failures in human-AI interactions
79% of AI failures in human interactions are invisible to users and form eight stable archetypes that persist even as models improve.