Z. Wang, Y. Ma, and C. Xu, “Eliciting Chain-of-Thought in Base LLMs via Gradient-Based Representation Optimization”, AAAI, vol. 40, no. 40, pp. 33782–33790, Mar. 2026.