QuantumMind Tests Quantum-Augmented LLM Reasoning
DeepTech reports that Zhao Junhua and his team are developing and commercializing QART (Quantum-Augmented Reasoning Transformer) through the quantum AI startup QuantumMind, where Zhao serves as chief scientist. QART combines a large language model with quantum optimization for some reasoning-path selection. The team says QART improved performance in 14 of 15 test groups across six benchmarks without retraining the base model, though gains varied by model and task. Zhao says adding quantum computing increases cost. The startup is exploring applications in power and finance, while the complete system is not open-sourced; more comparisons with classical optimization methods are needed to assess the quantum module’s contribution.
Long-form reasoning remains vulnerable to accumulated errors, and QART offers an optimization-based approach to selecting reasoning paths. However, its reported benefits are task-dependent, and its business value depends on whether performance gains justify the additional cost.
近年来,随着人工智能技术快速发展,量子计算与 AI 的结合也受到越来越多关注。
长期以来,量子计算被寄予厚望,希望借助不同于传统计算机的计算方式,在材料模拟、组合优化等领域解决一些复杂问题。如今,大模型的发展又为这一领域带来了新的研究方向。一方面,研究人员尝试利用 AI 改进量子计算的控制和算法设计;另一方面,也有人开始探索,量子计算能否反过来提升 AI 系统的能力。
其中一个受到关注的方向,是大模型的复杂推理。随着推理模型不断发展,AI 已经能够处理越来越复杂的数学、编程和科学问题,但在需要连续完成大量步骤的任务中,模型仍可能因为中途的判断错误而影响最终结果。如何提高长程推理的可靠性,仍是当前大模型研究面临的挑战之一。
#QuantumMind#QART#Boson Quantum#D-Wave#Guangdong Power Grid#Zhao Junhua