Yang Zhilin

From Tsinghua to Pittsburgh

Yang took his bachelor's degree at Tsinghua University in 2015, where Jie Tang advised him. He then went to Carnegie Mellon for a PhD, advised by Ruslan Salakhutdinov and William W. Cohen. It was finished in 2019, with a thesis titled Advances in Generative Feature Learning.712

His own site says he also worked at Meta AI with Jason Weston and at Google Brain with Quoc V. Le.7

Three papers in nine months

HotpotQA, a dataset for multi-hop question answering, came out in September 2018. Transformer-XL followed in January 2019. His thesis describes it as a design that models longer-range dependency.61112

XLNet arrived in June 2019 and uses a permutation language modelling objective. He wrote on his site that, as of September 2019, "Among my papers, my favorite ones as of Sept 2019 are XLNet, Transformer-XL, and HotpotQA". OpenAlex lists 217 works and 24,022 citations for the matching author record.571215

Moonshot and Kimi

Moonshot was established on 1 March 2023. Yang founded it with Zhou Xinyu and Wu Yuxin. It launched with a 100-billion-parameter model, and its chatbot Kimi was reported to take 200,000 Chinese characters in a single conversation.34

In a March 2024 interview Yang said a large-model company needs talent and capital in one place. He also said AI should be judged over ten to twenty years, not the next one or two.1

The Recurrent AI arbitration

Five investors in Recurrent AI, Yang's earlier company, filed a case against him at the Hong Kong International Arbitration Centre. A GSR Ventures representative alleged that Yang had uncleared liability when he launched Moonshot.213

Yang told the South China Morning Post: "I have gone through all necessary formalities to leave Recurrent and start a new business". He said Recurrent's board had approved his exit. His lawyers said the matter "lacks both legal grounds and factual support". In October 2026 Reuters reported that Moonshot executives were in settlement talks ahead of a planned Hong Kong IPO, and that the dispute was still in arbitration.2816

Accusations of copying US models

These claims are made against Moonshot as a company, not against Yang personally. Anthropic alleged in February 2026 that Moonshot used hundreds of fraudulent accounts to generate more than 3.4 million exchanges with Claude. In July 2026 White House adviser Michael Kratsios alleged that Moonshot distilled an Anthropic model to develop Kimi K3, but gave no details of how the government learned this.910

A Chinese embassy spokesperson told Reuters the claims were entirely unfounded, and Moonshot had not commented. In September 2026 a joint NSA, CISA and FBI advisory named Moonshot among firms it said ran distillation campaigns, with no penalty reported.1417

Yang, year by year

1. 2015

   Received his bachelor's degree from Tsinghua University, where Jie Tang advised him.7

2. Sep 2018

   Co-authored HotpotQA, a dataset for multi-hop question answering.11

3. Jan 2019

   Co-authored Transformer-XL, a language model built for longer context.612

4. Jun 2019

   Co-authored XLNet, which uses a permutation language modelling objective.512

5. 2019

   Finished his PhD at Carnegie Mellon, advised by Ruslan Salakhutdinov and William W. Cohen. His thesis was called Advances in Generative Feature Learning.712

6. Mar 2023

   Moonshot AI was established on 1 March, with Zhou Xinyu and Wu Yuxin as co-founders.34

7. Feb 2024

   TechCrunch reported that Moonshot had launched with a 100-billion-parameter model and raised more than $1 billion in a Series B.4

8. Dec 2024

   Five investors in Recurrent AI filed for arbitration against Yang in Hong Kong.213

9. Feb 2026

   Anthropic alleged that Moonshot used hundreds of fraudulent accounts to pull more than 3.4 million exchanges from Claude.9

10. Jul 2026

    White House science adviser Michael Kratsios alleged that Moonshot distilled an Anthropic model to build Kimi K3.10

11. Sep 2026

    A joint NSA, CISA and FBI advisory named Moonshot among Chinese firms it says ran distillation campaigns against US models.14

12. Oct 2026

    Reuters reported Moonshot executives in settlement talks over the Recurrent claim, ahead of a planned Hong Kong IPO.8

Questions people ask about Yang Zhilin

Yang Zhilin on Kenning

We haven't written about Yang Zhilin yet. New stories will appear here.

Moonshot AI on Kenning