This Guy Yao Shunyu Really Doesn't Hold Back, Bashing Anthropic, OpenAI, and Various 'Old Farts'

@Signal and Noise: I spent most of the day listening to Zhang Xiaojun's 4-hour long interview with Yao Shunyu.

This guy, who jumped from Anthropic to Google DeepMind just last year, participated in Claude 3.7/4.5 and Gemini 3. He provided a lot of genuine perspectives from a frontline researcher at a top-tier large model company.

The information density of the interview is incredibly high. Here are a few points I found most interesting:

1. Google prohibits employees from using Claude Code, but Yao Shunyu conservatively estimates that 90% of his code is AI-generated. A less conservative estimate would be 99% or even 100%. If a physics Ph.D. from Tsinghua and Stanford and a top-tier large model researcher relies on AI to write code, anyone else claiming they can't use AI to write code is just kidding themselves. On the flip side, though, if Google doesn't even allow CC and Codex internally, how much must employee efficiency be impacted? This company is truly bizarre.

2. Among his reasons for leaving Anthropic, opposition to Dario's anti-China stance accounted for 40%. He said this wasn't the primary reason, but it was definitely a major one. He is highly annoyed by Dario's logic that "we must possess the strongest model to push AI safety." There really aren't many people in the circle who dare to openly bash their former employer from just half a year ago.

3. The naming of Claude 3.5/3.6/3.7 was an amateurish blunder. Anthropic's early product capabilities were extremely weak—"they actually called two models by the same name." The outside world spontaneously started calling it 3.6 to distinguish them, and Anthropic later followed the community's convention by naming the next one 3.7. I always thought 3.6 was a skipped version number.

4. Claude Code is "the beginning of individual heroism." A researcher named Boris wanted to build an efficiency tool for himself, which later turned into one of Anthropic's most important products. It grew entirely from the bottom up, not from top-down planning.

5. Not a single person from Anthropic's founding team has left. The core group from OpenAI had fought battles together, and that is the root of why a top-down culture can work there. In contrast to OpenAI, where all the executives have left, Yao Shunyu seems to hold OpenAI's corporate culture and some of its executives in quite low regard.

6. OpenAI saved Google's life. The logic is quite contrarian: if ChatGPT had swallowed the search market in one go, Google would have been finished; precisely because OpenAI created the possibility but didn't push it to the absolute limit, it gave Google time to strike back.

7. The most important trait in the AI industry isn't brains, it's reliability. His exact words: "Those things just require more brains; undergraduates can do them." A physics Ph.D. saying this is essentially a dimensional strike, but it also serves as a reassurance to everyone wanting to pivot to AI.

8. He believes the future of programmers is 1/1000 of them earning 100 times the salary. It's not that "programmers will disappear," but rather "extreme centralization." The vast majority will lose their unique value, while a small minority at the top will make a fortune.

9. He thinks that when many people say the Scaling Law has hit a wall, it's mostly due to their own code bugs. His exact words: "The progress brought by fixing a single bug is far greater than some magical tricks." Pre-training has continued to grow stronger over the past few months, completely contradicting the external narrative that "pre-training is dead."

10. The vast majority of New Labs will die.

My biggest takeaway from listening to this: this guy really doesn't hold back, bashing Anthropic, OpenAI, and various "old farts." But the source of his confidence to speak out is quite clear—he's not on the SSI track, nor does he rely on LPs for a living. In his own words: "There are no mentors or old friends in my industry, so of course I can bash whoever I want."

Also, he mentioned that he probably won't stay at Google for too long either. Saying this out loud on a podcast makes me think that top-tier domestic large model companies should start fighting to recruit him.