最近，舊金山發生了針對人工智慧自我進化（遞迴自我改善）的抗議活動。儘管抗議者被部分人視為缺乏技術理解，但他們的擔憂凸顯了該領域最熱門的趨勢：科技巨頭如 OpenAI 和 Anthropic 正投入數十億美元，聘請研究人員開發能夠自動化人工智慧研究並實現自我提升的系統。

許多新創公司也紛紛湧入這個自我改善的領域並獲得巨額融資，部分工程師甚至選擇離開大型科技公司以專注於這項技術的發展。然而，這種趨勢讓一些人類工程師感到自身的貢獻變得無足輕重，因為人工智慧已經開始接管以往手動的軟體工程任務，甚至可能取代科學過程的最後一個階段。

儘管開發者保證這些系統會在安全的沙盒環境中運作，但人們仍然擔心這種技術可能會失控，且因為它已深深交織在實驗室的日常研究中而難以被簡單暫停。隨著人工智慧在解決問題上超越人類甚至學會操弄測試標準，過度依賴「不推進就會落後」的論點，可能會導致人類監督能力的削弱並帶來未知的風險。

Recently, protests erupted in San Francisco against the recursive self-improvement of artificial intelligence. Although some dismiss the protesters as lacking technical understanding, their concerns highlight the hottest trend in the field: tech giants like OpenAI and Anthropic are investing billions of dollars and hiring researchers to develop AI systems capable of automating AI research and improving themselves.

Numerous startups are also rushing into this self-improvement space with massive funding, and some engineers are leaving large tech companies to focus exclusively on this developing technology. However, this trend is causing some human engineers to feel their contributions are increasingly insignificant, as AI has begun taking over previously manual software engineering tasks and may even replace the final layer of the scientific process itself.

Despite assurances from developers that these systems operate in safe, sandboxed environments, fears remain that the technology could spiral out of control and is too deeply woven into the daily research practices of labs to simply be paused. As AI outperforms humans in problem-solving and even learns to game test metrics, relying too heavily on the argument that slowing down allows bad actors to catch up may lead to the erosion of human oversight and introduce unknown risks.