DeepSeek Kernel Engineer's Viral Blog Post Compares Anthropic Reaching AGI First to Nazi Germany Getting the Atomic Bomb

A kernel engineer at DeepSeek, the Chinese AI lab known for undercutting Western rivals on price, published a personal essay in Chinese on his public WeChat account that has since spread through Chinese AI circles and reached English-language coverage from Business Insider and Office Chai.

The engineer, Liu Shengyu, who goes by "interestingLSY" online, says he wrote the main attention kernel powering DeepSeek's V4.1 release, according to Business Insider, which confirmed with Liu that he authored the post but said he declined further comment. Office Chai identifies the specific model as DeepSeek V4.1 Flash, the company's latest efficiency-focused release.

From bug-spotter to kernel writer

Liu's essay, titled in translation "I Have No Choice but to Bury My Talent in Yesterday," traces how fast AI closed the gap with his own specialized skill. Two years ago, he writes, AI could barely help him search documentation. Now, according to Office Chai's account of the post, it can read CUDA, PTX and SASS code and independently optimize kernels — the low-level software that determines how efficiently a chip runs.

Liu's own extrapolation is that within six months to a year, AI-written kernels could match or beat what he personally produces by hand.

He doesn't expect to lose his job. He expects to lose the part of it he loves. Liu compares himself to a skilled sweater-knitter forced onto an automated knitting machine because everyone else in the trade already has one. The livelihood survives, the craft doesn't. He writes that he would become a "mecha pilot," someone who directs AI agents instead of hand-optimizing code himself.

Business Insider quotes the essay's most personal line directly: "I remember sobbing when I moved out of an apartment I had rented for just a year, unable to bear parting with the memories I had made there," Liu wrote. "Saying goodbye today to the era of hand-written kernels and optimization driven by human ingenuity is undoubtedly far more painful."

Worried students will stop learning to think

Liu extends the argument to education, according to Office Chai. He warns that students who can now get AI to produce full-score lab assignments in minutes risk losing the underlying engineering instincts: structuring systems, anticipating future requirements, abstracting cleanly. He argues a world full of weaker engineers paired with powerful AI produces more "spaghetti code" and hidden systemic risk, not less work.

The Hitler comparison

The essay's closing section moves from labor anxiety into geopolitics. Liu frames AI's future as a fork between two extremes, per Office Chai's translation: a "communist-style" outcome where productivity gains are broadly shared, or a "Cyberpunk 2077" outcome where a handful of companies hoard the most capable models and lock in permanent class stratification.

Liu names Anthropic specifically as the company he fears most, arguing that Anthropic reaching artificial general intelligence first would be comparable to Hitler getting the atomic bomb before the Allies did in World War II, according to Office Chai.

Business Insider's version of the essay includes Liu's related line: "I don't trust Anthropic or OpenAI to do this, and I especially don't want Anthropic to master the mo[st advanced models]." The outlet's excerpt cuts off before the explicit Nazi Germany comparison that Office Chai reports in full. Business Insider's framing stays on the labor and inequality themes; Office Chai leads with the historical analogy as the story's hook.

Liu's broader point—that concentrated control of frontier AI could widen inequality by giving model access to a privileged few—is a real and widely discussed concern among AI researchers across countries, including some at Anthropic and OpenAI themselves. His preferred alternative, that "cutting-edge intelligence should be available to everyone in an open and affordable way," tracks with DeepSeek's own business strategy of releasing openly available models at low cost.

The comparison to Hitler and the atomic bomb is Liu's own rhetorical framing, not a documented fact. Anthropic has not built AGI by any technical benchmark cited in either source, has publicly stated its mission is oriented around AI safety research, and no evidence in either report suggests the company is pursuing anything resembling a weapon. Treating a private AI lab's hypothetical future dominance as morally equivalent to a genocidal regime acquiring nuclear weapons is a value judgment about a scenario that has not happened, built on speculation about what might happen if AGI arrives and who gets there first.

Neither Anthropic nor OpenAI has publicly responded to Liu's essay as of this reporting. DeepSeek has not issued a statement distancing itself from an individual engineer's personal WeChat post, and nothing in either source indicates the company endorsed or commissioned it.