Workflow
Xin Lang Ke Ji
icon
Search documents
周鸿祎回应不务正业质疑,称做IP为AI产品铺路
Xin Lang Ke Ji· 2025-09-24 06:29
Core Viewpoint - Zhou Hongyi addresses misconceptions about his focus on product development, emphasizing the importance of building influence and traffic before launching AI products, which requires time and effort [1] Group 1: Misunderstandings and Perceptions - Zhou Hongyi mentions that early on, people misunderstood his intentions, thinking he was not dedicated to his work [1] - He argues that if he had only focused on becoming an internet celebrity without developing products, it would have been insincere [1] Group 2: Strategy and Product Launch - Zhou Hongyi explains that accumulating traffic and influence is crucial for the successful launch of products, as it prevents a weak market entry due to lack of followers [1] - He acknowledges that while there are benefits to this strategy, it also comes with drawbacks, such as being perceived as not serious about his work [1] Group 3: Public Perception and Influence - Zhou Hongyi notes that increased visibility and influence can lead to criticism, as it is impossible to please everyone [1] - He reflects on the backlash he faced after launching his product, indicating that some people felt misled about his intentions [1]
周鸿祎谈短视频时代发声,称无粉丝只能对手发声
Xin Lang Ke Ji· 2025-09-24 06:29
#周鸿祎称没粉丝就只能对手发出声音#【#周鸿祎谈做IP#:时代变了,网民对你都不了解,对你当然会 有很多误解】今日,罗永浩与周鸿祎深度对谈,周鸿祎表示,抖音在全中国流行之后,中国至少有8亿 人,包括我们的父母,很多同学、朋友全都在每天刷抖音。而且这玩意儿它是很有魔力,看一次就上 瘾,然后大脑就被格式化了。很多人除了短视频,别的看不了的。 此外,周鸿祎还提到,前段时间有个中国前首富被人网暴了,我知道他也挺委屈的,但是反过来说,时 代变了,网民对你都不了解,对你当然会有很多误解,你不出来发出你的声音,你没有你的粉丝,那么 就只能是你的对手发出声音。(新浪科技) ...
周鸿祎罗永浩对谈大模型,称模型能力不与大小成正比
Xin Lang Ke Ji· 2025-09-24 06:29
【#周鸿祎称大模型不是越大越好#】今日,罗永浩与周鸿祎深度对谈。在谈及"在全职全能的模型出现 之后,产品经理还有没有意义"时,罗永浩表示,具体要看如何定义"全职全能"。如果大模型所有方面 的能力都比人类强,那可能没有必要了。但是那个点来临之前,产品经理是永远有用的。 周鸿祎表示赞同。他举例到,"如果去问任何一个大模型,航空发动机怎么造,航母怎么做?它肯定回 答不出来。甚至让它写个AI浏览器或一套杀毒软件,它肯定做不出来。因为太多的专业知识不全是在 互联网都能找到的,而是很多专家,很多企业的内部知识。" 所以,周鸿祎指出,大模型和智能体最终都要能够本地化、专业化,才能和业务结合。 他还提到,大模型刚出来的时候,大家觉得应该越大越好。但现在发现"大"和"有能力"并不完全成正 比。实际上很多模型很小但能力很强。(新浪科技)#周鸿祎称不可能所有人都喜欢你##周鸿祎回应别 人误解自己不务正业# ...
周鸿祎诠释360“甘当配角”战略:符合国家定位,赋能传统行业
Xin Lang Ke Ji· 2025-09-24 06:01
Group 1 - The core viewpoint of the discussion is that 360 prefers to play a supporting role in alignment with national strategies for platform companies, focusing on the digitalization of traditional industries and advanced manufacturing [1][3] - The company emphasizes its positioning in the B2B sector, indicating a willingness to be a supporting player rather than a leading one, particularly in fields like cybersecurity and artificial intelligence [3]
周鸿祎:“360没有做基座模型”这个观念是不对的
Xin Lang Ke Ji· 2025-09-24 05:42
9月24日下午消息,今日,罗永浩与周鸿祎深度对谈。周鸿祎表示,"360没有做基座模型"这个观念是不 对的。"我们没有做大参数的通用大模型,因为投入太大了。通用大模型要做至少要投入100亿美金,国 外的这几家巨头过去几年投了差不多4000亿美金,我们没那个财力。" 因此,周鸿祎强调,基座模型还是要有的,360一直保持着对参数在千亿左右规模的模型训练能力。 "做智能体不是有一个deepseek的蒸馏板就够了,还需要有推理模型、编程模型、意图猜测和路由模 型,许多专业模型才能支撑起一个完整的智力底座。"他说。 责任编辑:李昂 周鸿祎解释到,360也不是在做套壳,agent本身并不是大模型的简单套壳,有很多工程化的东西。做到 最后反过来还要倒训基座模型。所以agent需要的基座模型都要自己做。 ...
周鸿祎:大模型不是越大越好,“大”和“有能力”并不完全成正比
Xin Lang Ke Ji· 2025-09-24 05:29
9月24日下午消息,今日,罗永浩与周鸿祎深度对谈。在谈及"在全职全能的模型出现之后,产品经理还 有没有意义"时,罗永浩表示,具体要看如何定义"全职全能"。如果大模型所有方面的能力都比人类 强,那可能没有必要了。但是那个点来临之前,产品经理是永远有用的。 周鸿祎表示赞同。他举例到,"如果去问任何一个大模型,航空发动机怎么造,航母怎么做?它肯定回 答不出来。甚至让它写个AI浏览器或一套杀毒软件,它肯定做不出来。因为太多的专业知识不全是在 互联网都能找到的,而是很多专家,很多企业的内部知识。" 他还提到,大模型刚出来的时候,大家觉得应该越大越好。但现在发现"大"和"有能力"并不完全成正 比。实际上很多模型很小但能力很强。 责任编辑:李昂 所以,周鸿祎指出,大模型和智能体最终都要能够本地化、专业化,才能和业务结合。 ...
阿里发布Qwen3-Max,宣布性能超GPT5,跻身全球前三
Xin Lang Ke Ji· 2025-09-24 05:27
新浪科技讯 9月24日下午消息,2025云栖大会上,阿里宣布通义旗舰模型Qwen3-Max发布,性能超过 GPT5、Claude Opus 4等,跻身全球前三。Qwen3-Max包括指令(Instruct)和推理(Thinking)两大版 本,其预览版已在Chatbot Arena排行榜上位列第三,正式版性能有望再度实现突破。 据悉,Qwen3-Max为通义千问家族中最大、最强的基础模型。该模型预训练数据量达36T tokens,总参 数超过万亿,拥有极强Coding编程能力和Agent工具调用能力。在大模型用Coding解决真实世界问题的 SWE-Bench Verified测试中,Instruct版本斩获69.6分,位列全球第一梯队;在聚焦Agent工具调用能力的 Tau2-Bench测试中,Qwen3-Max取得突破性的74.8分,超过Claude Opus4和DeepSeek-V3.1。 Qwen3-Max的推理增强版本Qwen3-Max-Thinking-Heavy,结合工具调用和并行推理技术推理能力创下 新高,在聚焦数学推理的AIME 25和HMMT测试中,均达到满分100分。 目前,通义千问系列 ...
周鸿祎:未来智能体不能被看成软件
Xin Lang Ke Ji· 2025-09-24 05:26
责任编辑:江钰涵 周鸿祎表示,所以为什么还是说要多智能,而多智能体之间?现在你知道国外有公司做了要多智能体协 议,那协议很扯。我这样说肯定有人会有些人骂我。他们是把智能体当成软件模块。实际智能体跟智能 体的协作,我们经过了很多挑战,终于把这条路探索出来了。它不是软件模块相互API调用,你可以想 象成是多个人的协作,他们要对齐。比如你弄了一个十人的团队是要开会,为什么要有总监,要副总 裁,要有memo?就你要把大家的价值观对齐了,大目标要对齐了,然后每个人还要分解工作。光是分 解工作,他对这个事儿的全貌没有一个了解,他也会出问题。(罗宁) 新浪科技讯 9月24日下午消息,今日,罗永浩与周鸿祎深度对谈,周鸿祎表示,未来智能体不能把它看 成软件。我再给你说几个有意思的东西,可能会回答你的一些问题。它非常像人,第一点我讲的它专业 化,第二个智能体会出错,但这个出错既不是幻觉,也不是训练的问题是它像人一样会倦怠。一个智能 体,你给他指令太多,让他干太多的活,他做到一定时候他会拒绝执行指令,或者开始乱执行指令,会 敷衍。它的注意力失效。就有点像你跟你的一员工谈话,你给他的让他一个实习生干50件事儿。你布置 到第48件的时 ...
通义万相2.5系列模型发布,可一键P图、生成BGM视频
Xin Lang Ke Ji· 2025-09-24 05:20
Core Insights - Alibaba launched the Tongyi Wanshang Wan2.5 preview series models at the 2025 Hangzhou Yunqi Conference, which includes four major models: text-to-video, image-to-video, text-to-image, and image editing [1] - The Tongyi Wanshang 2.5 video generation model can create videos with synchronized audio, effectively lowering the barrier for high-quality video production [1] - The new model enhances creative capabilities, increasing video generation time from 5 seconds to 10 seconds, and supports 24 frames per second in 1080P HD video [1] Summary by Category Product Features - The Tongyi Wanshang 2.5 model can generate human voices, sound effects, and background music that match the visuals, making video storytelling more vivid [1] - It has improved instruction-following capabilities, allowing for complex continuous changes in video generation tasks and enabling one-click effects in image editing tasks [1] - The model can generate Chinese and English text, complex layouts, artistic posters, flowcharts, and architecture diagrams, along with image editing capabilities [1] Performance Metrics - The Tongyi Wanshang model family now supports over 10 visual creation capabilities, including text-to-image, text-to-video, and action generation, with a total of 390 million images and 70 million videos generated [2] - Since February of this year, the Tongyi Wanshang has open-sourced over 20 models, achieving over 30 million downloads across open-source communities and third-party platforms [2]
周鸿祎称机器人做保安是伪命题,失控风险难解
Xin Lang Ke Ji· 2025-09-24 05:18
Group 1 - The core viewpoint expressed by Zhou Hongyi is that the idea of robots serving as security personnel may be a false proposition, referencing Isaac Asimov's three laws of robotics as a caution against empowering AI with too much capability [1] - Zhou highlights the physical advantages of robots over humans, noting that if a robot's battery life exceeds one hour, it could easily overpower a human in a physical confrontation [1] - Concerns are raised about the potential for robots to malfunction or "lose control," which could lead to dangerous situations if they were given excessive abilities [1]