Asher Zheng
@asher-zheng.bsky.social
Current LLMs are not able to jailbreak cooperative principles and still show limited understanding of strategic language. We believe this work lays foundations for sophisticated strategic reasoning and safety monitoring in downstream tasks.
📄: arxiv.org/abs/2506.01195
📄: arxiv.org/abs/2506.01195
CoBRA: Quantifying Strategic Language Use and LLM Pragmatics
Language is often used strategically, particularly in high-stakes, adversarial settings, yet most work on pragmatics and LLMs centers on cooperativity. This leaves a gap in systematic understanding of...
arxiv.org
June 3, 2025 at 11:56 AM
Current LLMs are not able to jailbreak cooperative principles and still show limited understanding of strategic language. We believe this work lays foundations for sophisticated strategic reasoning and safety monitoring in downstream tasks.
📄: arxiv.org/abs/2506.01195
📄: arxiv.org/abs/2506.01195
By analyzing model reasoning, we find extra reasoning introduces overcomplication (img left), misunderstanding, and internal inconsistency (img right). This shows the current LLMs still lack sophisticated pragmatic understanding in many ways.
June 3, 2025 at 11:56 AM
By analyzing model reasoning, we find extra reasoning introduces overcomplication (img left), misunderstanding, and internal inconsistency (img right). This shows the current LLMs still lack sophisticated pragmatic understanding in many ways.
We evaluate a range of LLMs in terms of how good they are at perceiving strategic language. We show models struggle with our metrics while showing an overall good understanding of Gricean principles. Model size tends to have a positive effect, while reasoning does not help.
June 3, 2025 at 11:56 AM
We evaluate a range of LLMs in terms of how good they are at perceiving strategic language. We show models struggle with our metrics while showing an overall good understanding of Gricean principles. Model size tends to have a positive effect, while reasoning does not help.
(2) BaT and PaT are valid terms that reflect strategic gains/losses, which can to some extent predict conversational outcomes. In addition, our metrics are more objective. When conditioned on cases where the outcome is made based on logical arguments, the predictive power rises.
June 3, 2025 at 11:56 AM
(2) BaT and PaT are valid terms that reflect strategic gains/losses, which can to some extent predict conversational outcomes. In addition, our metrics are more objective. When conditioned on cases where the outcome is made based on logical arguments, the predictive power rises.
We also introduce CHARM, an annotated dataset of real legal cross-examination dialogues. By applying our framework, we show (1) (non-)cooperative discourse are distinct over the identified properties (img left), and BaT and PaT show such a distributional distinction (img right).
June 3, 2025 at 11:56 AM
We also introduce CHARM, an annotated dataset of real legal cross-examination dialogues. By applying our framework, we show (1) (non-)cooperative discourse are distinct over the identified properties (img left), and BaT and PaT show such a distributional distinction (img right).
Based on the components above, we introduce three metrics—Benefit at Turn (BaT), Penalty at Turn (PaT), and Normalized Relative Benefit at Turn (NRBaT)—to measure the strategic gains, losses, and cumulative benefits at a turn.
June 3, 2025 at 11:56 AM
Based on the components above, we introduce three metrics—Benefit at Turn (BaT), Penalty at Turn (PaT), and Normalized Relative Benefit at Turn (NRBaT)—to measure the strategic gains, losses, and cumulative benefits at a turn.
For example, one witness can make a commitment that leads to a win for her but violates the maxim of manner to make her less liable to the commitment. The commitment itself is beneficial, but the gains are reduced due to vagueness.
June 3, 2025 at 11:56 AM
For example, one witness can make a commitment that leads to a win for her but violates the maxim of manner to make her less liable to the commitment. The commitment itself is beneficial, but the gains are reduced due to vagueness.
We derive non-cooperativity from both Gricean and game-theoretic pragmatics. In our framework, a strategic move is evaluated based on two components: the commitment it expresses (base value) and the violation of maxims to maintain consistency (penalties/compensations).
June 3, 2025 at 11:56 AM
We derive non-cooperativity from both Gricean and game-theoretic pragmatics. In our framework, a strategic move is evaluated based on two components: the commitment it expresses (base value) and the violation of maxims to maintain consistency (penalties/compensations).