Frontier LLMs show a strong bias toward literal communication in subtext tasks, generating overly explicit clues 60% of the time in visual allusion games, with partial improvement only when common ground is provided.
They can see the patterns you create over time
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Beneath the Surface: Investigating LLMs' Capabilities for Communicating with Subtext
Frontier LLMs show a strong bias toward literal communication in subtext tasks, generating overly explicit clues 60% of the time in visual allusion games, with partial improvement only when common ground is provided.