谷歌论文提前揭示o1模型原理：AI大模型竞争或转向硬件

澎湃新闻09-18

OpenAI最强模型o1的护城河已经没有了?仅在OpenAI发布最新推理模型o1几日之后，海外社交平台 Reddit 上有网友发帖称谷歌Deepmind在 8 月发表的一篇论文内容与o1模型原理几乎一致，OpenAI的护城河不复存在。澎湃科技（www.thepaper.cn)注意到，谷歌DeepMind团队于今年8月6日发布上述论文，题为《优化 LLM 测试时计算比扩大模型参数规模更高效》（...

网页链接

免责声明：本文观点仅代表作者个人观点，不构成本平台的投资建议，本平台不对文章信息准确性、完整性和及时性做出任何保证，亦不对因使用或信赖文章信息引发的任何损失承担责任。

精彩评论

我们需要你的真知灼见来填补这片空白

发表看法

{"i18n":{"language":"zh_CN"},"isChannel":false,"data":{"share":"https://www.laohu8.com/m/news/2468848355?lang=zh_CN&edition=full","thumbnail":"","is_english":false,"pubTime":"2024-09-18 21:37","share_image_url":"https://static.laohu8.com/9a95c1376e76363c1401fee7d3717173","id":"2468848355","market":"us","top_or_hot":-1,"title":"谷歌论文提前揭示o1模型原理：AI大模型竞争或转向硬件","media":"澎湃新闻","content":"<div>\n<p>OpenAI最强模型o1的护城河已经没有了?仅在OpenAI发布最新推理模型o1几日之后，海外社交平台 Reddit 上有网友发帖称谷歌Deepmind在 8 月发表的一篇论文内容与o1模型原理几乎一致，OpenAI的护城河不复存在。澎湃科技（www.thepaper.cn)注意到，谷歌DeepMind团队于今年8月6日发布上述论文，题为《优化 LLM 测试时计算比扩大模型参数规模更高效》（...</p>\n\n<a href=\"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20240918213745956d2687&s=b\">网页链接</a>\n\n</div>\n","source":"tencent","html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>谷歌论文提前揭示o1模型原理：AI大模型竞争或转向硬件</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 11px; color: #7E829C; margin: 0;line-height: 11px;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\n谷歌论文提前揭示o1模型原理：AI大模型竞争或转向硬件\n</h2>\n\n<h4 class=\"meta\">\n\n\n2024-09-18 21:37 北京时间&nbsp;&nbsp;&nbsp;<a href=http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20240918213745956d2687&s=b><strong>澎湃新闻</strong></a>\n\n\n</h4>\n\n</header>\n<article>\n<div>\n<p>OpenAI最强模型o1的护城河已经没有了?仅在OpenAI发布最新推理模型o1几日之后，海外社交平台 Reddit 上有网友发帖称谷歌Deepmind在 8 月发表的一篇论文内容与o1模型原理几乎一致，OpenAI的护城河不复存在。澎湃科技（www.thepaper.cn)注意到，谷歌DeepMind团队于今年8月6日发布上述论文，题为《优化 LLM 测试时计算比扩大模型参数规模更高效》（...</p>\n\n<a href=\"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20240918213745956d2687&s=b\">网页链接</a>\n\n</div>\n\n\n</article>\n</div>\n</body>\n</html>\n","isBrief":false,"type":0,"news_type":1,"symbol":"LU0128525689.USD","symbol_name":"TEMPLETON GLOBAL BALANCED \"A\"(USD) ACC","start_time":0,"source_url":"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20240918213745956d2687&s=b","article_id":"2468848355","we_media_id":null,"thumbnails":[],"rights":{"source":"tencent","url":"http://gu.qq.com/resources/shy/news/detail-v2/index.html#/?id=nesSN20240918213745956d2687&s=b","rn_cache_url":null,"customStyle":"body{padding-top:10px;}#news_title{font-weight:bold;#titleStyle#;}#news_description span{font-size:12px;#descriptionStyle#;}.footer-note{#statement#}","selectors":".mod-LoadTzbdNews, body","filters":".relate-stock, .hot-list, .recom-box, .wx-sou","directOrigin":true},"url":"https://stock-news.laohu8.com/highlight/detail?id=2468848355","pubTimestamp":1726666620,"sourceInfo":{"source_id":"tencent","name":"腾讯"},"weMediaInfo":null,"summary":"仅在OpenAI发布最新推理模型o1几日之后，海外社交平台 Reddit 上有网友发帖称谷歌Deepmind在 8 月发表的一篇论文内容与o1模型原理几乎一致，OpenAI的护城河不复存在。该论文探讨了“思维链”技术，旨在突破 Transformer 模型在串行推理方面的限制。OpenAI 近期发布的o1 模型，或正是上述理念的实践。目前，英伟达在AI大模型算力的分配上占据主导地位。这些迹象表明，大模型的竞争已不仅局限于模型本身，硬件能力也成为关键因素。","collect":0,"end_time":0,"defaultTopTitle":"qq.com","property":[],"viewcount":null,"language":"zh","relate_stocks":{"LU0128525689.USD":"TEMPLETON GLOBAL BALANCED \"A\"(USD) ACC","IE00B5TLWC47.USD":"BNY MELLON LONG-TERM GLOBAL EQUITY \"B\" (USD) ACC","IE00BSNM7G36.USD":"NEUBERGER BERMAN SYSTEMATIC GLOBAL SUSTAINABLE VALUE \"A\" (USD) ACC","LU0130518102.USD":"HARRIS ASSOCIATES GLOBAL EQUITY \"R\" INC","NVDA":"英伟达","BK4554":"元宇宙及AR概念","BK4532":"文艺复兴科技持仓","LU0203202063.USD":"AB SICAV I - ALL MARKET INCOME PORTFOLIO \"A2X\" (USD) ACC","AMD":"美国超微公司","BK4553":"喜马拉雅资本持仓","MSFT":"微软","LU0061474705.USD":"THREADNEEDLE (LUX) GLOBAL DYNAMIC REAL RETURN \"AU\" (USD) ACC","IE00BWXC8680.SGD":"PINEBRIDGE US LARGE CAP RESEARCH ENHANCED \"A5\" (SGD) ACC","LU0097036916.USD":"贝莱德美国增长A2 USD","LU0215105999.USD":"SCHRODER ISF GLOBAL EQUITY \"A\" ACC","BK4533":"AQR资本管理(全球第二大对冲基金)","LU0106261372.USD":"SCHRODER ISF US LARGE CAP \"A\" ACC","BK4566":"资本集团","IE00B1BXHZ80.USD":"Legg Mason ClearBridge - US Appreciation A Acc USD","BK4587":"ChatGPT概念","LU0225283273.USD":"SCHRODER ISF GLOBAL EQUITY ALPHA \"A\" (USD) ACC","LU0061475181.USD":"THREADNEEDLE (LUX) AMERICAN \"AU\" (USD) ACC","IE0034235188.USD":"PINEBRIDGE GLOBAL FOCUS EQUITY \"A\" (USD) ACC","BK4077":"互动媒体与服务","LU0149725797.USD":"汇丰美国股市经济规模基金","BK4538":"云计算","BK4579":"人工智能","LU0127658192.USD":"EASTSPRING INVESTMENTS GLOBAL TECHNOLOGY \"A\" (USD) ACC","IE00BJTD4N35.SGD":"Neuberger Berman US Long Short Equity A1  Acc SGD-H","LU0029864427.USD":"TEMPLETON GLOBAL \"A\" (USD) INC","LU0211328371.USD":"TEMPLETON GLOBAL EQUITY INCOME \"A\" (MDIS) (USD)   INC","LU0433182093.SGD":"First Eagle Amundi International AS-C SGD","GOOG":"谷歌","IE0004086264.USD":"BNY MELLON GLOBAL OPPORTUNITIES \"A\" (USD) ACC","BK4573":"虚拟现实","GOOGL":"谷歌A","BK4561":"索罗斯持仓","IE00B3S45H60.SGD":"Neuberger Berman US Multicap Opportunities A Acc SGD-H","LU0068578508.USD":"First Eagle Amundi International Cl AU-C USD","IE00BJJMRX11.SGD":"Janus Henderson Balanced A Acc SGD","BK4581":"高盛持仓","LU0456855351.SGD":"JPMorgan Funds - Global Equity A (acc) SGD","LU0079474960.USD":"联博美国增长基金A","LU0082616367.USD":"摩根大通美国科技A（dist）","LU0056508442.USD":"贝莱德世界科技基金A2","LU0640476718.USD":"THREADNEEDLE (LUX) US CONTRARIAN CORE EQ \"AU\" (USD) ACC","LU0080751232.USD":"富达环球多元动力基金A","LU0130517989.USD":"HARRIS ASSOCIATES US VALUE EQUITY \"R\" INC","LU0719512351.SGD":"JPMorgan Funds - US Technology A (acc) SGD","IE00BFXG0V08.USD":"BNY MELLON GLOBAL LEADERS \"B\" (USD) ACC","BK4548":"巴美列捷福持仓","LU0308772762.SGD":"Blackrock Global Allocation A2 SGD-H","IE00BKDWB100.SGD":"PINEBRIDGE US LARGE CAP RESEARCH ENHANCED \"A5H\" (SGDHDG) ACC"},"translate_title":"Google paper reveals the principle of o1 model in advance: AI large model competition may turn to hardware","themeId":null,"isJumpTheme":false,"ttsUrl":null,"symbols_score_info":{"GOOGL":0.9,"GOOG":0.9,"NVDA":0.7,"AMD":0.7,"MSFT":0.6},"content_text":"OpenAI最强模型o1的护城河已经没有了?仅在OpenAI发布最新推理模型o1几日之后，海外社交平台 Reddit 上有网友发帖称谷歌Deepmind在 8 月发表的一篇论文内容与o1模型原理几乎一致，OpenAI的护城河不复存在。澎湃科技（www.thepaper.cn)注意到，谷歌DeepMind团队于今年8月6日发布上述论文，题为《优化 LLM 测试时计算比扩大模型参数规模更高效》（Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters）。谷歌DeepMind团队于今年8月6日发布的论文在这篇论文中，研究团队探讨了大模型（LLM）在面对复杂问题时，是否可以通过增加测试时的计算量来提高决策质量。这项研究表明，增加测试时（test-time compute）计算比扩展模型参数更有效。基于论文提出的计算最优（compute-optimal）测试时计算扩展策略，规模较小的基础模型在一些任务上可以超越一个14倍大的模型。无独有偶，另一篇由谷歌和斯坦福大学研究人员于今年1月发表的论文《思维链赋能 Transformer 解决本质上的串行问题》（Chain of Thought Empowers Transformers to Solve Inherently Serial Problems）也提出了类似的观点。该论文探讨了“思维链”（Chain of Thought，简称 CoT）技术，旨在突破 Transformer 模型在串行推理方面的限制。传统的Transformer模型擅长并行计算，但在处理需要逻辑推理的复杂问题时表现欠佳。CoT的核心思想是让模型模拟人类的思考方式，通过生成一系列中间推理步骤，来解决复杂问题。OpenAI 近期发布的o1 模型，或正是上述理念的实践。o1模型在给出答案之前，会生成一系列中间推理步骤，不断完善自己的思维过程，尝试不同的策略，并能识别自身错误。随着更多的强化学习和思考时间，o1的性能持续提升。有网友表示，“所有的秘密突破和算法最终都会随着顶尖开发者在行业内的流动而传播到其他公司和开源社区。”谷歌也表示没有人拥有护城河，这也促使OpenAI将o1-mini的速度提高7倍，每天都能使用50条；o1-preview则提高每周50条。有网友评论道：“唯一可能形成护城河的是硬件，至少在可预见的未来是这样。”也有人认为，如果AI大模型公司无法解决对显存的依赖，英伟达可能会直接掌控谁能够获得计算能力。而如果微软或谷歌开发出在自研芯片上运行速度快10倍的模型，情况也会变化。目前，英伟达在AI大模型算力的分配上占据主导地位。值得注意的是，OpenAI近期也被曝出其首款芯片计划，采用台积电最先进的A16级工艺，专为Sora视频应用打造。这些迹象表明，大模型的竞争已不仅局限于模型本身，硬件能力也成为关键因素。在AI领域，谁能拥有更强大的算力，谁就可能在下一阶段的竞争中占据优势。","kind":null,"is_publish_news":true,"is_publish_highlight":false,"is_publish_live":false,"is_publish_wemedia":null,"editions":null,"column":"","sentiment":"0","news_tag":"Product Release","news_rank":0,"symbols":[],"gpt_button":0},"commentList":[],"isCommentEnd":true,"newsSizeData":{"likeSize":0,"commentSize":0,"repostSize":0,"favoriteSize":0,"likeStatus":false,"favoriteStatus":false},"APP":{"userAgent":"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)","isDev":false,"isTTM":false,"deviceId":"web-server-community-laohu8-v3","version":"4.28.0","shortVersion":"4.28.0","platform":"web","vendor":"web","appName":"laohu8","isIOS":false,"isAndroid":false,"isTiger":false,"isTHS":false,"isWeiXin":false,"isWeiXinMini":false,"isWeiBo":false,"isQQ":false,"isBaiduSwan":false,"isBaiduBox":false,"isDingTalk":false,"isToutiao":false,"isOnePlus":false,"isHuaWei":false,"isXiaomi":false,"isXiaomiWebView":false,"isOppo":false,"isVivo":false,"isSamsung":false,"isMobile":false},"href":"/m/news/2468848355"}