首页 > 最新小说 > 女帝

正义的伙伴

Gemini掉出全球前十,谷歌新模型为何没能赢得掌声?_我的网站

爸爸去哪儿

A |     

Ministry of Commerce
    Ministry of Commerce
China on Monday decided to impose anti-dumping tariffs of up to 54.3 percent on imports of pecans (Carya illinoensis) from Mexico and the US in the form of deposits after it ruled preliminarily that imports of pecans from the two countries were dumped in the Chinese market, China's Ministry of Commerce (MOFCOM) announced.  
The investigating authority found a causal link between the alleged dumping of imports of pecans from Mexico and the US and the injury suffered by Chinese producers. The ministry said it will impose provisional anti-dumping measures in the form of security deposits beginning August 11, 2026, according to an announcement on the ministry's website on Monday.
The dumping margins for Mexican companies were determined to be 17.8 percent to 51.6 percent. Since no US company has responded to the investigation, and in accordance with Chinese law and WTO rules, the dumping margin for all US companies was determined on the basis of the best information available at 54.3 percent, according to a spokesperson of the MOFCOM. 
China launched the anti-dumping investigation into pecan imports from Mexico and the US on September 25, 2025, under its Anti-Dumping Regulations, according to the ministry.
A number of Mexican companies responded to the investigation, but no US companies did so. 
China has always exercised caution and restraint in using trade remedy measures, and remains firmly committed to safeguarding fair and free trade, the MOFCOM spokesperson said on Monday, briefing on the preliminary ruling. China will continue to conduct the investigation in accordance with the law, fully protect the rights of all interested parties, and make an objective and impartial final ruling based on the findings of the investigation, the spokesperson said.
Pecans, also known as American walnuts or thin-shelled walnuts, are used both as snacks and in baking, and can also be processed into edible oil.
The MOFCOM spokesperson said that the ministry has conducted the investigation in accordance with the principles of fairness, impartiality, openness, and transparency, and in strict compliance with relevant Chinese laws and regulations as well as WTO rules.
The ministry said interested parties may submit written comments within 10 days of the announcement.
Global Times
。    当地时间7月21日,谷歌DeepMind发布三款轻量的新模型,主打“更快、更智能、更具成本效益”,但因智能程度没有满足外界的期待,评价两极分化。         与此同时,旗舰模型Gemini 3.5 Pro依然没有露面成为此次焦点,从排行榜来看,谷歌的模型已经排在十名开外。         根据谷歌官方博客,此次发布的三款模型定位不同。         Gemini 3.6 Flash是主力模型,与上一代相比能使用更少的token,在相同成本下提供更高质量的工作;而Gemini 3.5 Flash-Lite则是快速、成本效益高的选项,适用于日常任务,如处理文档和搜索;Gemini 3.5 Flash Cyber是一种网络安全模型,专为发现和修补关键软件漏洞而构建。         但根据评测来看,此次谷歌的新模型在智能上并不出色。AI模型评测平台Arena.ai的研究显示,Gemini 3.6 Flash 在前端代码竞技场中以1537分排名第12位。         第三方评测机构Artificial Analysis的结论则是,Gemini 3.6 Flash 在智能方面较3.5 Flash没有提升,智能指数得分50,但任务效率有所提高,Gemini 3.6 Flash的平均每任务时间为 1.3 分钟,较上一代减少超过50%,同时每任务成本略有降低。         这意味着,谷歌优化的更多是速度和成本,而不是模型能力本身。         更值得关注的是,在Artificial Analysis最新智能指数排行榜中,谷歌目前已经没有任何一款模型进入前十。排名靠前的是Anthropic、OpenAI等海外厂商,也包括国内月之暗面的Kimi K3、智谱GLM-5.2等模型。这与去年Gemini一度站上全球第一梯队形成了鲜明反差。         除了排行榜,开发者社区的反馈同样出现明显分化。不少开发者用谷歌新模型做任务时未看到很好的表现,“前端能力糟糕、空间推理也很差,我安慰自己这其实不是在高思考水平上运行的,但我不确定。”          有开发者评论道,“谷歌就像龟兔赛跑里的兔子,领先一段时间后就开始睡觉。”也有人直言,Gemini 团队需要引入真正关心交付前沿模型的人。         在关于谷歌新模型发布的评论中,相比Flash模型,大家更关注的是,旗舰模型Gemini 3.5 Pro什么时候发布?          在5月的谷歌I/O 大会上,谷歌发布了Gemini 3.5 Flash,当时曾预告Pro版本已在内部使用,预期一个月内推出。然而两个月过去了,Gemini 3.5 Pro依然没有公开亮相。据外媒报道,由于模型始终未达到内部预期性能,谷歌推迟了发布计划。         面对外界质疑,谷歌DeepMind技术负责人洛根·基尔帕特里克(Logan Kilpatrick)也在社交媒体上回应称,Gemini 3.5 Pro正在与合作伙伴进行测试,计划在准备就绪后立即广泛提供。但他未给出具体的时间节点。         与此同时,谷歌在这次博客中提到,团队已经开始着手构建下一代模型,“启动了迄今为止规模最大的 Gemini 4 预训练工作,并对目前的进展感到振奋”。这一表态,也让不少业内人士猜测,谷歌是否正在把更多资源转向下一代模型,而非继续打磨3.5 Pro。         针对现阶段谷歌模型的表现,很多人认为,并不是谷歌的能力不行,而是内部的策略方向问题,优先级放在产品上,而不是像Anthropic、OpenAI等厂商专注将AI做好。         从谷歌这次的更新来看,团队在大模型上的策略似乎是“速度胜过智慧”。有开发者认为,让任务时间缩短是很好,但停滞的智能意味着他们优先考虑了错误的指标。“创新也需要深度,而不仅仅是秒表”。

B |          不过,也有开发者认为,如果模型是为了赚钱,从这个商业角度看,谷歌是聪明的,毕竟现在新兴公司都是在烧钱运营,一旦钱烧完直接就倒闭,而谷歌是将模型降智节省成本。         相比Anthropic和OpenAI持续围绕模型能力竞争,谷歌拥有Android、Chrome、Workspace、搜索、云服务等庞大的产品生态,因此这一年其AI战略一直更偏向产品落地,而不是单纯追逐模型能力和榜单。         但另一方面,开发者更倾向于将智能作为评价AI公司的第一标准。如果模型能力长期停滞,即便商业化领先,也难免影响开发者生态和市场预期。         遥想去年11月,谷歌发布Gemini 3模型时曾引发业界震动,反超OpenAI、坐稳第一梯队,股价一路飙升。

C | 而进入2026年以来,AI竞争的风向再次发生变化。OpenAI依靠持续推出新模型重新夺回行业关注度,Anthropic依旧保持着模型能力领先的位置,而谷歌更多展示的是成本优化和产品整合能力。         截至7月21日收盘,谷歌(GOOGL.US)跌超1%,股价收于347.15美元/股,较5月中旬高点累计回落近15%,当前谷歌总市值约4.2万亿美元,也从全球市值榜第二退居第三,被苹果赶超。         (本文来自第一财经)。

Current article:http://7u2.senchuoshaozaodidia.shop/news/20260826_745066.html

Published on:14:35:57