Meta continues to plunder all AI labs that they can find as they look to develop a superintelligence team of its own.
Meta has poached three Google DeepMind researchers who were involved in the company’s model which won a gold medal at the International Mathematics Olympiad, The Information reports. This comes after Meta had poached at least 10 top OpenAI researchers and several researchers from other labs to build its Superintelligence Unit led by Alexandr Wang. Google DeepMind’s model had become the first AI system to officially win a gold medal at the International Mathematics Olympiad.
The 3 researchers which Meta has reportedly poached from Google DeepMind are are Tianhe Yu, Cosmo Du and Weiyue Wang. Tianhe Yu was a student researcher at Google before joining Google DeepMind in 2022. He worked on Gemini RL, Thinking and post-training. He drove the Gemini 2.5 Pro launch and was a core contributor to Gemini 2.5 Pro Deep Think. Cosmo Du was a Principal Scientist & Director at Google DeepMind working on Gemini Post-training, Thinking, and Coding. He was a core contributor of Gemini 1, 1.5, 2, and 2.5. Weiyue Wang had worked at Waymo for nearly 5 years, and had been at Google DeepMind since 2024. She had been one of the main contributors of Gemini 2.5 pro and Gemini 2.0 thinking.
In recent months, Meta has been on a hiring spree, looking to hire top talent from AI labs. Meta’s own AI efforts were widely perceived to be lagging behind the top labs after the failure of its Llama 4 model, which not only delivered underwhelming performance, but was quickly upstaged by Chinese open-source models. This appears to have prompted Meta to go into war mode, and the company is doing all that it can to lay its hands on the top AI talent. Meta has reportedly been offering $100 million signing bonuses and $300 million pay packages over 4 years to pull away top talent from other labs. Most of Meta’s early efforts had been focused on OpenAI, but now seem to be extending to other labs.
This is possible in Silicon Valley because companies in the area famously don’t have non-compete agreements, and employees can join rivals almost immediately. This worked well in the PC and mobile eras, where individual researchers didn’t have all that much domain specific knowledge that they could take to other companies. But in AI, where researchers could have the specific know-how to build models that are worth tens of billions, this model now seems to being stretched to its maximum.