Tian Keyu

Headlines · 4

1. 1d
   Tian Keyu's stealth AI lab raises ~$30M at $200M valuation, drawing criticism

   According to Bloomberg, a roughly 10-person AI lab with no product and no public name has raised nearly $30 million at a $200 million post-money valuation, backed by 5Y Capital and IDG Capital, and led by founder Tian Keyu, an early-20s Peking University PhD student. Tian is first author of the NeurIPS 2024 best paper (VAR) and was the subject of ByteDance's alleged "model poisoning" incident: a first-instance court ordered him to pay 500,000 yuan and return all internship salary while rejecting ByteDance's 8 million yuan tort claim, and ByteDance has appealed with the second trial still pending. 36Kr's commentary argues that, with strong AI projects heavily oversubscribed, investors who "close their eyes" are effectively pricing controversy-driven attention as an asset.

   36氪 人工智能 · 🔥 4
2. 3d
   Tian Keyu's stealth world-model startup raises ~$30M as poisoning allegations resurface

   A Chinese long-form commentary examines the case of Tian Keyu, a former ByteDance intern whom the company fired in August 2024 and later accused of tampering with colleagues' model-training code; ByteDance said in an October 2024 statement that reports of 8,000+ GPUs and tens of millions of dollars in losses were heavily exaggerated. A GitHub page posted the same day as his podcast appearance, signed by five named people including co-first authors of the video project Goku and the first author of LlamaGen, alleged the interference began in March 2024 and included corrupted text-encoder weights, a modified optimizer, altered data shuffling and a shared-storage backdoor; the page also claimed the VAR text-to-image project served PixArt, an open-source model, for daily tests. Tian, who won a NeurIPS best-paper award for VAR, has since raised close to $30 million from 5Y Capital, IDG and others at a roughly $200 million post-money valuation for a still-unnamed world-model company of about ten people; a court reportedly found him in breach of contract and ordered 500,000 yuan in damages.

   虎嗅 AI · 🔥 0
3. 4d
   Explainer: What world models are, and how they differ from LLMs

   An explainer republished by 36Kr from the WeChat account Xianzao Ketang walks through the surging concept of “world models”: rather than replicating a full digital world, they aim to give AI an internal, runnable simulator of physical reality that internalizes gravity, collisions, occlusion and causality as底层 capabilities—here rendered in English as core abilities. The article argues that today's large language models rely on statistical patterns and “memorize answers,” knowing facts but not the world, while world models emphasize common-sense reasoning, forecasting and cross-domain transfer. It cites embodied AI and autonomous driving as early uses—generating rare edge cases and letting robots trial actions in a “dream”—while flagging extremely high training costs, weak handling of complex social interactions and a long road to large-scale commercialization. The funding and acquisition figures it cites (AMI Labs' seed round, AMD's purchase of World Labs, Tian Keyu's new startup) come only from that article and were not independently verified here.

   36氪 人工智能 · 🔥 0
4. 6d
   Ex-ByteDance intern Tian Keyu raises nearly $30M for world-model startup

   Former ByteDance intern Tian Keyu has started a "world model" company that has raised several tens of millions of dollars — nearly $30 million — at a post-money valuation of about $200 million (roughly 1.4 billion yuan), according to a 36Kr affiliate report, with 5Y Capital partner Meng Xing incubating and leading the investment. Tian had been sued by ByteDance over interfering with internal model training during his 2024 internship; foreign media citing the verdict say the court ordered him to pay ByteDance 500,000 yuan for breach of contract and forfeit all of his salary. Tian says the roughly 10-person startup has built an AI "dictionary" of 200,000 symbols to help models process video, plans to train on 100 million hours of video, and aims to show off the technology at live events before releasing a full model in 2027.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my Tian Keyu experienceAsk about Tian Keyu