Textual-gradient methods automate prompt optimization through natural-language feedback, but their iterative updates can be unstable. We identify two sources of this instability: noisy gradients produced from already-correct examples and over-specialization to hard cases that degrades performance on simpler inputs. We...
Yi-Fan Xu, Yi-Xuan Li, Xin-Zhuo Li et al.· 0 citations
The results indicate that typed state transitions and deterministic evidence control contribute beyond fluent generation alone on quantum-acceleration hypotheses beyond fluent generation alone.
Yijing Zuo, Zhengkang Fu, Zihan Nie et al.· 0 citations
Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks. However, existing approaches suffer from two key challenges: brittle multi-step and multi-turn reasoning caused by incompatible tool output types and API schemas, and performance degra...
Hai-Bo Jin, Sui-Jin Wang, Xuchen Yu et al.· 0 citations
A synthetic player population is constructed whose traits are ground truth by construction, and an opportunity-aware decision-moment representation is introduced that disentangles preference from the chance to express it; ablating it selectively degrades opportunity-dependent traits.
