Yao Shunyu Hires for Tencent Hunyuan AI4S with a 50-Year Math Problem Solved

Yao Shunyu, a prominent AI researcher, has personally started recruiting for Tencent's Hunyuan AI4S team. Instead of a detailed job description, he posted a simple message with a research result: their self-developed agent Hyra has advanced a 50-year-old open problem in mathematics to a new best result. Hyra, built on the Hy3 model (295B total parameters, 21B active), found a construction that proves the upper bound in a problem about sums and differences of integer sets can be approached arbitrarily closely. The agent first searched for better results, then switched to natural language to propose a mathematical construction and proof, running for about 24 hours. The team independently verified and formalized the proof in Lean 4. In initial AI4S tests, Hyra-1.0 achieved impressive results across mathematics, astronomy, quantum computing, and drug design, including solving 29 of 55 open problems, discovering a predictive formula from sunspot data (R²=0.77), improving a quantum routing algorithm by 44.4%, and generating a PARP1 candidate molecule. The team aims to achieve full automation in AI research, where AI proposes hypotheses, runs experiments, and iterates. To that end, they seek both AI researchers (for agent architecture, RL, context learning, evaluation, training systems, GPU kernels) and interdisciplinary scientists who can translate scientific goals into executable tasks and judge the validity of AI results. This trend reflects a broader shift where AI is becoming a true researcher, as evidenced by a popular comment that this may be the last Fields Medal awarded to purely human work. Notably, OpenAI and Anthropic have jointly called for international mechanisms to coordinate AI development speed, highlighting the dual reality of AI self-improvement and the need for control.