Re: [討論] NV發布完全不用人類審查代碼的專案

看板Soft_Job作者 (3d)時間3小時前 (2026/02/17 07:37), 編輯推噓0(000)
留言0則, 0人參與, 最新討論串3/3 (看更多)
我以為你是收錢辦事的網軍,但現在知道你是帶來歡樂的小丑。現在ai公司在系統灑了太多錢在宣傳 vibetensor有論文 https://arxiv.org/abs/2601.16238 後面有講到問題,我稍微節錄一下 "The “Frankenstein” composition effect. A recurring failure mode in generated systems is that individually reasonable components can compose into a globally suboptimal design." 就是架構有問題。 "Incomplete API surface and performance. VibeTensor intentionally does not aim for full PyTorch compatibility. Many operators, datatypes, and distributed features are missing or incomplete, and performance has not been tuned to match production frameworks." 功能有限,性能有問題。就是toy. "Validation gaps unique to generated code. Agent-generated code can pass local unit tests while failing under repeated composition (e.g., multi-step training loops) due to stateful interactions, uninitialized buffers, or accidental global synchronization." bugs很多,小玩一下ok,真跑起來會當機。 "Maintenance, safety, and security. Machine-generated code can include inconsistent conventions, redundant abstractions, and subtle correctness or security issues. We therefore caution against production use and position VibeTensor primarily as a research and educational artifact." 程式碼問題很多,當研究課題ok,但不要當真。 最近很多這類東東,但出來的結果都跟這篇差不多。哪天心情比較好,比較有空再閒聊目前ai寫程式的問題與應用。 ※ 引述《yamakazi (大安吳彥祖)》之銘言: : https://github.com/NVlabs/vibetensor : 詳細內容可以問AI : NVDA 發表了一個完全由AI寫的專案 : 人類只給高階提示詞 : 完全不寫一行代碼 : 不審核一行代碼 : 當然跟原本PyTorch 比效能差了一些 : 不過只花了兩個月完成 : ----- : Sent from JPTT on my iPhone -- ※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 58.114.66.74 (臺灣) ※ 文章網址: https://www.ptt.cc/bbs/Soft_Job/M.1771285031.A.C51.html
文章代碼(AID): #1fawednH (Soft_Job)
文章代碼(AID): #1fawednH (Soft_Job)