1 paper · 1 filter
Yunbo Long, Haolang Zhao, Lukas Beckenbauer +2
Post-trained LLMs are often optimized to align responses with human preferences, making them safe, polite, and conversationally appropriate. In adversarial negotiation, however, th…