@jinchenma_ai: I watched Xiaojun Zhang's interview with Yao Shunyu for 4 hours, packed with valuable insights. He made a particularly contrarian judgment. Many say pre-training has hit a wall and Scaling Law has reached its limit. He says no, and there are no signs of hitting a ceiling in the coming months. So why do so many people think it's hit a wall? He directly said: the vast majority of people who shout about hitting a wall have bugs in their own code...
Summary
In the interview, Yao Shunyu proposed a contrarian view that pre-training has not hit a wall and Scaling Law has not reached its limit, claiming that most people who say it has hit a wall have bugs in their code.
View Cached Full Text
Cached at: 05/22/26, 01:55 PM
I watched that interview with Yao Shunyu by Zhang Xiaojun—4 hours long, extremely dense with干货.
He had one judgment that was particularly contrarian.
Many people say pre-training has hit a wall or Scaling Law has reached its limit. He said no, and there’s no sign of hitting a ceiling in the coming months either.
So why do so many people think it’s hitting a wall? He said bluntly: most of those claiming it’s hitting a wall simply have bugs in their own code and aren’t aware of it.
Fixing one bug often brings more progress than all those fancy tricks put together.
Similar Articles
@AlchainHust: Spent most of a day listening to the 4-hour interview between Zhang Xiaojun and Yao Shunyu. This guy, who just moved from Anthropic to Google DeepMind last year, has worked on Claude 3.7/4.5 and Gemini 3. He offered many candid perspectives from a frontline researcher at a top large model lab. The interview is incredibly information-dense…
This article summarizes Zhang Xiaojun's interview with Yao Shunyu, a researcher involved in developing Claude and Gemini, who shares 10 insightful views on AI code generation, company culture, scaling laws, etc.
@Fenng: Such a despicable scumbag can still get interviewed by the media and fawned over by investors. No wonder even Sun Ge wa…
该推文评论了字节跳动前实习生田柯宇涉嫌攻击模型训练事件，批评其仍能接受媒体采访并获得投资人追捧，并引用了受害者陈述与证据整理的帖子。
@sheriyuo: I think @latepostnews's reporting is indisputably the best in the Chinese-language region, whether it's the earlier DeepSeek exclusive, Top Seed, or this latest big-package narrative. They really dare to write—Xin Zhi Yuan / Quantum Bit and the like can't even come close, and as for Heart of Machine, it's just distilling a few tweets on Twitter these days (...
User @sheriyuo praises LatePost as the best AI media in the Chinese-language region, criticizes competitors like Xin Zhi Yuan, Quantum Bit, and Heart of Machine, and mentions the depth of reports on DeepSeek and others.
@Bill_Do_A_Bit: https://x.com/Bill_Do_A_Bit/status/2056581340842066212
Based on Yao Shunyu's analysis, the article contends that AI will prioritize transforming tasks that have clear feedback loops and quick validation, rather than by job prestige. Programmers are the first to be impacted because of the comprehensive testing and feedback mechanisms inherent in code development. Although a product manager's core decision-making is hard to train, their peripheral execution layers are also headed for disruption.
@huoru_tarnish: Zhang Yiming is the real master of palace intrigue. Many people still tout his "loves technology, engineer persona," but the reality is completely different.
This tweet argues that Zhang Yiming is not simply an engineer who loves technology, but a master of office politics who secured his position through a series of power plays, citing the development of Douyin and Volcano as examples.