WizardLM team left Microsoft for Tencent after scandal with model closure

Tencent, known for developing the WeChat messenger and popular games, has invited the WizardLM research group, previously working under Microsoft’s wing, to join its team. And the first results of the collaboration are already visible. The team presented the Hunyuan-TurboS 0416 model. According to Qingfeng Sun, who calls himself one of the founders of WizardLM, this development surpasses open artificial intelligence models, including the Gemma 3 series from Google.

And I’m telling you about this transition because the history of the WizardLM research group is notable for one ambiguous episode. In April of last year, still under the auspices of Microsoft, the team released a family of WizardLM-2 models, claiming that they were not inferior in capabilities to GPT-4 from OpenAI. However, just a day later, Microsoft removed WizardLM-2 from public access, explaining this by the lack of toxicity testing.

The situation even drew criticism from Clement Delangue, CEO of the Hugging Face platform for artificial intelligence developers. He noted that Microsoft’s actions damaged the community, disrupting the work of many open-source projects. You understand? Microsoft is the largest investor in OpenAI. And I think they didn’t want anyone to beat the OpenAI model, in which many billions of dollars have been invested.

At Tencent, the WizardLM team will likely continue to do the same thing as before – developing and releasing artificial intelligence models. Recently, the Chinese company just created 2 new divisions and increased spending on artificial intelligence infrastructure to 12 billion dollars.

And the story with WizardLM demonstrates that Chinese technology giants can attract already formed research groups with experience working in Western companies. Which contributes to reducing the technological gap with the USA.