iPhone hacker George Hotz: AI coding agents are the coming disaster

MarketWhisper

AI編碼代理

Famous hacker George Hotz (who first cracked the iPhone at age 17, had reverse-engineered the PlayStation 3 before Sony sued him, and is the author of the open-source deep learning framework Tinygrad) published a blog post on May 25 titled 《The Eternal Sloptember》, claiming that “applying AI agents to software development will be one of the most costly mistakes in the history of the field.

核心論點:高低績效團隊的不對稱輸出問題

Hotz 的論點建立在組織結構差異上:高績效團隊擁有足夠緊密的回饋機制,能夠發現代理生成的問題;低績效團隊缺乏自我檢查機制,卻「利用代理實現了比以往高出十倍的產出」。在大型科技公司中,他預測結果是「程式碼品質平均水準的快速下降,而這種下降卻被龐大的程式碼量所掩蓋」,最終導致「大量劣質產品氾濫成災,優質產品黯然失色」。他舉例:報導稱蘋果正在其整個工程團隊中推廣 AI 編碼工具,他直接問道:「你認為未來兩年 macOS 會變得更好還是更糟?」

六個月測試的具體結論與 Hotz 的立場歸屬

Hotz 表示在測試中,代理「會預先完成所有工作,遞給你一個老虎機拉桿——你拉動它,然後祈禱剩下的工作能夠順利完成。它始終無法完全實現。」他嘗試了讓模型自動處理代碼、工程師審核輸出的模式,但表示「每次都必須手動修改」。他將自己歸入 Meta 首席 AI 科學家 Yann LeCun 和 AI 批評者 Gary Marcus 的陣營,認為語言模型本質上是複雜的模式匹配器,無法解決真正的新問題。

業界分歧:Karpathy 和 Amodei 的對立立場

Andrej Karpathy 曾在 2025 年初對代理持懷疑態度,但在新模型發布後改變立場,於 5 月 19 日加入 Anthropic 預訓練團隊,稱接下來幾年的工作為「意義非凡的成長經驗」。Dario Amodei 在達沃斯表示,部分 Anthropic 工程師已停止編寫代碼而讓模型處理。微軟 CEO Satya Nadella 也將 GitHub Copilot 向代理系統的轉型定義為平台層面的變革。

常見問題

Is Hotz’s argument aimed at all engineers using AI tools?

No. Hotz is explicit that the issue is not being replaced, but the scale effect of organizations: high-performing engineers can spot problems produced by agents, while low-performing engineers cannot. His criticism is aimed at the downside effect on overall code quality when low-performing teams use agents at scale.

Hotz 提到的「越來越難以察覺」的問題具體指什麼?

Hotz 表示「輸出結果存在問題,但這種問題越來越難以察覺,這正是我們對一個日益精確的統計模型所預期的」。問題不在於代理產生明顯的錯誤,而在於產生外表正確但內在存在缺陷的代碼,缺乏足夠能力的工程師無法識別這些問題。

What do the differences between Hotz and Karpathy represent?

These two engineers with strong industry credibility hold opposing views: Hotz believes AI coding agents are a disaster, while Karpathy (now joined at Anthropic) believes agents have fundamentally changed software development. Their disagreement reflects that the industry has not yet reached a conclusion on the real benefits of AI coding agents.

Disclaimer: The information on this page may come from third-party sources and is for reference only. It does not represent the views or opinions of Gate and does not constitute any financial, investment, or legal advice. Virtual asset trading involves high risk. Please do not rely solely on the information on this page when making decisions. For details, see the Disclaimer.
Comment
0/400
No comments