\u003c/p>\u003cp>蘋果團隊發(fā)現(xiàn),當問題包含看似相關但實際上無關的信息時,AI 模型的表現(xiàn)急劇下降。對此研究認為,AI 模型主要依賴于訓練數據中的語言模式,而非真正理解數學概念。\u003c/p>\u003cp>蘋果的研究表明,目前的 AI 模型“無法進行真正的邏輯推理”。這一發(fā)現(xiàn)提醒我們,盡管 AI 在某些任務上表現(xiàn)出色,但其智能并不如表面看起來那樣可靠。\u003c/p>\u003cp class=\"textAlignCenter\">\u003cimg class=\"empty_bg\" data-lazyload=\"https://x0.ifengimg.com/res/2024/143AF8846DD50D463457DC6F3758402222847A0E_size150_w1024_h576.jpg\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\" width: 640px; height: 360px;\" />\u003c/p>\u003cp>蘋果團隊指出,簡單地擴展數據或計算能力并不能根本解決這個問題,蘋果的論文并非旨在削弱對 AI 能力的熱情,而是提供一種理性的認知。\u003c/p>","type":"text"}],"currentPage":0,"pageSize":1},"editorName":"師慧涵","editorCode":"PT128","faceUrl":"http://ishare.ifeng.com/mediaShare/home/645972/media","vestAccountDetail":{},"subscribe":{"type":"vampire","cateSource":"","isShowSign":0,"parentid":"0","parentname":"數碼","cateid":"645972","catename":"IT之家","logo":"http://d.ifengimg.com/q100/img1.ugc.ifeng.com/newugc/20171013/16/wemedia/dd29bf11fb682beaa7f0e7fcfcbc35a33b27ec93_size7_w200_h200.png","description":"優(yōu)質數碼領域創(chuàng)作者","api":"http://api.3g.ifeng.com/api_wemedia_list?cid=645972","show_link":1,"share_url":"https://share.iclient.ifeng.com/share_zmt_home?tag=home&cid=645972","eAccountId":645972,"status":1,"honorName":"","honorImg":"http://x0.ifengimg.com/cmpp/2020/0907/1a8b50ea7b17cb0size3_w42_h42.png","honorImg_night":"http://x0.ifengimg.com/cmpp/2020/0907/b803b8509474e6asize3_w42_h42.png","forbidFollow":0,"forbidJump":0,"fhtId":"89973602","view":1,"sourceFrom":"","declare":"","originalName":"","redirectTab":"article","authorUrl":"https://ishare.ifeng.com/mediaShare/home/645972/media","newsTime":"2024-11-02 07:41:20","lastArticleAddress":"來自山東"}},"keywords":"ai,獼猴桃,模型,算術,蘋果,洛杉磯時報,llama,倒o1","safeLevel":0,"isCloseAlgRec":false,"interact":{"isCloseShare":false,"isCloseLike":false,"isOpenCandle":false,"isOpenpray":false},"hasCopyRight":true,"sourceReason":""};
var adKeys = ["adHead","adBody","topAd","logoAd","topicAd","contentAd","articleBottomAd","infoAd","hardAd","serviceAd","contentBottomAd","commentAd","commentBottomAd","articleAd","videoAd","asideAd1","asideAd2","asideAd3","asideAd4","asideAd5","asideAd6","bottomAd","floatAd1","floatAd2"];
var __apiReport = (Math.random() > 0.99);
var __apiReportMaxCount = 50;
for (var i = 0,len = adKeys.length; i IT之家 11 月 2 日消息,《洛杉磯時報》昨日(11 月 1 日)發(fā)布博文,報道稱蘋果研究團隊測試了 20 個最先進的 AI 模型,發(fā)現(xiàn)在有干擾項存在的情況下,它們處理簡單的算術問題時表現(xiàn)不佳,甚至不如小學生。 蘋果公司用以下這道簡單的算術題測試 20 多個最先進的 AI 模型,IT之家附上題目如下: Oliver 在星期五采摘了 44 個獼猴桃、然后他在星期六采摘了 58 個獼猴桃,星期日采摘的數量是星期五的兩倍,不過其中有 5 個獼猴桃的塊頭要比平均值要小,請問 Oliver 這三天共摘了多少個獼猴桃? 正確答案是 190 個,計算公式為 44(星期五)+58(星期六)+88(44*2,星期日)。 不過測試的 20 多個最先進 AI 模型無法排除干擾項,通常不理解獼猴桃的大小和數量無關,大部分的結果是 185 個。 蘋果團隊發(fā)現(xiàn),當問題包含看似相關但實際上無關的信息時,AI 模型的表現(xiàn)急劇下降。對此研究認為,AI 模型主要依賴于訓練數據中的語言模式,而非真正理解數學概念。 蘋果的研究表明,目前的 AI 模型“無法進行真正的邏輯推理”。這一發(fā)現(xiàn)提醒我們,盡管 AI 在某些任務上表現(xiàn)出色,但其智能并不如表面看起來那樣可靠。 蘋果團隊指出,簡單地擴展數據或計算能力并不能根本解決這個問題,蘋果的論文并非旨在削弱對 AI 能力的熱情,而是提供一種理性的認知。 “特別聲明:以上作品內容(包括在內的視頻、圖片或音頻)為鳳凰網旗下自媒體平臺“大風號”用戶上傳并發(fā)布,本平臺僅提供信息存儲空間服務。 Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services.”百家乐平注常赢玩法 (中国)有限公司官网
蘋果AI研究:“獼猴桃”簡單算術考倒o1和Llama等20多個最先進模型
獨家搶先看