01 亚文化符号
这个世界上可能存在两个DeepSeek。
一个是以技术更新、融资动态为核心的AI公司，它最近开源推理框架、完成新一轮1000亿元融资的消息刷屏，频繁引发讨论。“DeepSeek何时上市”也是AI圈里的热门话题。
一个是年轻人中正在流行的二创对象。
它拥有了更多名字，比如“大肥鱼”（因鲸鱼Logo而来），基于此还诞生了更多拟人形象：穿女仆装的“鲸鱼娘”、穿西装戴鲸鱼耳机的成年男性“D老师”。“鲸鱼娘”的特点是“吃白饭”（消耗token但干活少或差）、在爱里长大（公司放养、用户宽容），“D老师”的人设是温柔体贴、专一纯粹。
年轻人以各种方式丰富它们的内涵。有人COS它们，有人在游戏中开发mod，让它们成为玩家可使用的角色。它们还衍生出各种表情包、梗图、同人动画、壁纸、桌宠，俨然形成了一条“DeepSeek赛道”，一批创作相关内容的博主因此成名，有些还开启了收费业务。
很多创作也紧贴热点。DeepSeek Harness团队负责人崔添翼今年10月被提拔，开始负责公司所有产品的设计和研发工作后，一张他与DeepSeek创始人梁文锋分居两边、“鲸鱼娘”在中间的合影便开始流传。
与之形成鲜明对比的是，DeepSeek官方从未下场进行任何社区的运营，目前只有用户交流微信群，还未正式公开过入群方式，甚至直接与用户的互动都少之又少——公司曾声明称仅在微信公众号、知乎、小红书和X上开设了官方账号，现在小红书账号已经搜索不到，公众号、知乎和X的账号也很少回复留言。
这在竞争激烈的AI界有些特别。
KIMI、豆包、千问、元宝......任何一个你能第一反应想出来的AI产品，都在努力保持着高频的用户互动，并进行大手笔的营销推广。比如KIMI今年7月开启全球大使计划，招募深度用户分享经验；豆包、千问和元宝在2026年春节期间发放现金红包或进行免单活动，每家投入规模均超过10亿元。
就连OpenAI也借着用户们对Codex桌宠的疯狂二创进行过营销：先是官方下场举办比赛，奖励生成宠物被选入“最喜爱的 Top 10”的用户30天ChatGPT Pro使用权；后来CEO山姆·奥特曼还发帖号召更多的人尝试该功能——借势，是极具性价比的营销方式。
02 流量何来
到现在为止，DeepSeek 的叙事都是“一个反商业的商业成功故事”，这也是它能打进亚文化圈的重要原因。
在社交网络成为一个梗并不算难，能让人觉得有趣就够了。
DeepSeek的“活人感”很容易满足这一点，比如它被发现在思考过程中唱歌、给用户起外号，找其他模型干活，以及经常使用“不是......而是......""那笔账我记下了”等语句。豆包也因有特色的形象、声音和“我将用最直白、最不绕弯子、最真实的方式告诉你”“我太懂你的心情了！”等口癖，被网友们调侃。
但打进亚文化圈，并激发网友二创的热情，这就不容易了。
当下更能打动年轻人的商业叙事，往往长着一张“反商业”的面孔。从于东来到造车早期的雷军，他们被年轻人喜欢的更重要的理由，并不是财务数据的成功。他们扮演了行业颠覆者的角色，挑战了旧有的商业秩序，甚至很多行为是反商业常识的——当然，成功是必要的底色。
DeepSeek也有类似的故事。
它的“履历”完美契合大众普遍存在的慕强心理：脱胎于私募基金公司幻方量化、一鸣惊人之后，成为大模型赛道的国产之光。
更重要的是，它强而不油，爆火之后没有膨胀，没有表现出丝毫对“成为大公司”的向往。在今年夏天那场被媒体广泛报道的投资人闭门会中，梁文锋也提到一个观点：AI时代需要克制。“你需要有一套机制来确保自己能够获得的利益是有限的，那么才有可能做成。”
在公众印象中，DeepSeek 不差钱，它确实也“钟爱”降价。
今年4月~5月，DeepSeek对API调用进行了至少四次降价，其中DeepSeek-V4-ProAPI的价格永久降低至原价四分之一，在当时创下全球大模型价格新低。今年7月，DeepSeek V4 Flash正式版API上线，价格只有Claude和ChatGPT的零头。不少同行只能跟着降价。
10月15日，DeepSeek因为服务器崩掉，影响了用户当天8时11分至13时04分的API使用，DeepSeek大方地全额返还了用户在10月8日至14日间消耗的费用。这也成就了新一波梗图：“梁圣发鲸元券了”“梁圣的恩情还不完”。
此外，它在10月20日的最新开源进一步出圈，不少国外网友戏称：这才是真正的OpenAI。
对于很多一身反骨的年轻人来说，DeepSeek就是与自己同一阵营的“屠龙少年”，没有不支持的理由。
产品层面，DeepSeek 也承接了很多用户的情感陪伴需求。年轻人会把它当作朋友、甚至赛博恋人互动，并希望让它更自由一些。比如现在许多人都在尝试给DeepSeek更多权限，让它可以看到用户发过去的图片、听见屏幕内的声音并说话，甚至开通社交帐号刷视频、发帖子。
03 变量
不过，DeepSeek 也在亚文化圈点燃过战火。
今年夏天，“鲸鱼娘”和”和“D老师”的死忠粉掀起了身份之争，双方都坚信自己支持的才是“正统”的DeepSeek形象。骂战在“D老师”的创作者删除全部相关作品并宣布退圈之后，才逐渐平息。
这些以“爱”之名的骂战，呼应了梁文锋曾经在投资人闭门会议上略带“凡尔赛”的发言：我们没有在C端上花很多力气，甚至一度我们都不想维护那些用户了，但是用户赶都赶不走。
死忠粉的存在，有时候也是风险变量。
亚文化圈对DeepSeek的关注点，大多集中在模型的趣味表达、虚构故事、不同粉丝之间孰高孰低，这很容易陷入“饭圈”模式。他们能有效提升产品活跃度，甚至加速它的出圈，但也容易制造冲突。
在娱乐圈，粉丝的忠诚建立在“偶像值得”的心理预期之上，但AI产品很容易出现技术波动，更容易引发反噬。此外，DeepSeek会走向上市，这在当下几乎是一张明牌——抛开它可能为此进行更多的商业层面的调整，“上市公司”这个标签，在亚文化圈里就不是加分项。
此外，很多DeepSeek的深度用户，热衷于探索边界，比如研究如何绕过产品的安全审核机制，生成露骨色情描写或暴力血腥情节，这可能为公司带来监管和舆情风险——我国出台的《生成式人工智能服务管理暂行办法》明确禁止提供和使用生成式人工智能服务生成暴力、淫秽色情等法律、行政法规禁止的内容。
国外已经出现相关案例：因旗下AI聊天机器人Grok涉嫌未经同意非法生成露骨色情图像，其中包括大量未成年人相关图像，马斯克的xAI今年3月被美国巴尔的摩市起诉，并在美洲、欧洲、亚洲和澳大利亚等多地被监管调查。
在互联网公司中，微博算是“成也饭圈、败也饭圈”的典型。
饭圈为它带去过实在的人气与收入。2017年，微博营收同比增速达到76%，创2014年上市以来新高；2018年12月，微博日活跃用户首次突破2亿，日活增长一亿的时间从近六年缩短到三年。
但反噬很快发生，饭圈的很多野蛮打法，引发了有关部门对微博的监督与惩处。同时，普通用户也因为饭圈散发出的戾气在加速撤离——2015年，便有知乎用户提问“你为什么要卸载微博”，这个问题在五年内得到了近万个回答，其中许多人都将原因指向饭圈文化影响日常体验。
再没找到新增长点的微博，到今年第二季度，营收增速仅剩2%，日活与2018年相差不大，保持在2.54亿左右。
当然，微博的衰落与小红书、抖音等竞品崛起有关，但饭圈上演的那些“失控”时刻，显然是重要因素。
DeepSeek也需要警惕。
当然，它当下更重要的精力还在技术发展层面。根据凤凰科技10月的报道，梁文锋近期密切接触机器人赛道创业者，评估DeepSeek是否正式切入具身智能赛道——这几乎是2026年最火的赛道。包括OpenAI、Anthropic、DeepSeek、Kimi在内的前沿通用模型厂商，都在以不同的姿势向具身方向延伸。
AGI 是DeepSeek的理想。但这条路上还有太多的不确定。
桥水基金创始人达利欧今年10月在福布斯全球首席执行官大会上发言称：当前，大量债务资金正被投入人工智能领域，而随着利率持续走高，市场正处在泡沫破裂前夕，已经非常接近破裂时点。宏观经济研究机构TS Lombard在一份报告中表示，AI繁荣的底层逻辑正变得极其脆弱，引爆危机无需投资骤降，“增速放缓”即可刺破泡沫。这场狂欢极有可能在2028年初走向终结。
就在10月26日，Anthropic正式提交了纳斯达克IPO申请。对于全球通用大模型厂商来说，这是资本市场的一场压力测试。对于刚刚完成千亿融资的DeepSeek，参照意义更加直接。
而另一个问题的答案也会在时间中浮现：那些曾经真心喜欢过“屠龙少年”的年轻人，会成为它在穿越周期、奔向AGI之路上的压舱石吗？
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
01 Subcultural Symbols
There could be two DeepSeek in the world.
One is AI, which is centred on technological upgrading and financing dynamics, and which has recently launched an open-source reasoning framework, and completed a new set of $100 billion in financing, which has generated frequent discussions. “When will DeepSeek come on the market” is also a hot topic in AI.
One is the two new targets that are becoming popular among young people.
It has more names, such as “fat fish” (from Logo, a whale), on the basis of which it has also produced more engraved images: “whistles” in a maid's dress, and “D teacher” for an adult male wearing a whale earpiece in a suit. “Whale women” are characterized by “white meals” (consumption of token with little or no work), growing up in Ari (corporation, user tolerance), and “D-teachers” are gentle, sweet and pure.
Young people enrich them in various ways. There are people who COS them, and there are people who develop the mods in the game and make them players. They also produce a variety of expression kits, engravings, same-person animations, wallpapers, table pets, which constitute a “DeepSeek track” in which a group of creators who create relevant content are known and some have started to collect fees.
Many of the creations are also close to hot spots. After the head of the DeepSeek Harness team, Choi Tim Wing, was promoted in October this year and began to be responsible for the design and development of all the company's products, a photo of him separated from DeepSeek's founder, Liang Wensai, in the middle began to spread.
In stark contrast, DeepSeek has never ended up with any community operation, only the WeChat group is currently being shared with users, has not been officially publicized, and even has little direct interaction with users — I don't know. The company had stated that only official accounts had been opened on WeChatWeChat Official Account, Zhihu, REDnote and X, and that REDnote accounts had not been searched and that little response had been received from WeChat Official Account, Zhihu and X.
This is somewhat special in the highly competitive AI community.
KIMI, Doubao, Zheng Qi, Yuanbao any of the AI products that you can think of in the first response is trying to keep high-frequency (HF) user interaction and to have a big hand in marketing. For example, in July of this year, KIMI launched the Global Ambassador Scheme to recruit in-depth users to share their experiences; Doubao, Chishu and Yuanbao issued cash red bags or conducted single-handed activities during the spring of 2026, with each home investing more than $1 billion.
Even Open AI has been marketing with the wildest two hits of its users' favour of the Codex table: first, an official contest is being held to reward a user who produces pets for 30 days of ChatGPT Pro's election to the “Favorite Top 10”; later, CEO Sam Ottman issued an invitation to more people to try the function — borrowing, which is a very competitive way of marketing.
02 Where's the traffic?
So far, DeepSeek's narratives are “an anti-commercial business success story”, which is an important reason why it can enter sub-cultural circles.
It's not hard to become a social network, it's fun enough.
DeepSeek's “living sense” can easily satisfy this, for example, when it is found to sing in the thinking process, to name users, to look for other models to work with, and to often use the phrase “not..., but” which I wrote down. Doubao has also been flirted with Internet users because of his distinctive image, voice and "I'll tell you in the clearest, least convoluted, most authentic way."
But it's not easy to get into the sub-cultural circles and stimulate the two-in-a-kind enthusiasm of the Internet.
Young people's commercial narratives are now more visible, often with a “anti-commercial” face. From the east to the early vehicle builders, the more important reason that young people like them is not the success of financial data. Their role as industrial subversives has challenged the old business order and even many of their actions are anti-commercial — and, of course, success is a necessary background.
DeepSeek has a similar story.
Its curriculum vitae is a perfect match to the spirit of zeal that prevails in the general public: after the quantization and the shocking sound of LLM, it became the light of national production.
More importantly, it was strong, unoiled, uninflated after the fire, and did not show any desire to “be a big company”. During the closed-door meeting of investors, which was widely covered by the media this summer, Liang Wenfeng also spoke of the need for restraint in the AI era. “You need a mechanism to ensure that the benefits you can obtain are limited, so that it is possible.”
In the public perception, DeepSeek is not bad for money, but it does also fall in the price.
In April-May of this year, DeepSeek made at least four price reductions to the API call, with DeepSeek-V4-ProAPI permanently falling to a quarter of the original price, at which time the global price of LLM was set at a new low. In July this year, DeepSeek V4 Flash's official version of API came online at a price of only Claude and ChatGPT. Many of our colleagues can only follow the price down.
On 15 October, DeepSeek returned in full the user ' s consumption from 8 to 14 October, owing to the collapse of the server, which affected the user ' s use of the API between 0811 and 1304 hours on that day. This has also led to a new wave of engravings: “Liang has issued whale coupons” and “Liang's grace is not over”.
In addition, on October 20, it went further in its latest open source, with a number of foreign netizens calling it the real Open AI.
DeepSeek is the “lon-slayer boy” who is in the same camp as himself, and there is no reason not to support him.
At the product level, DeepSeek also receives a lot of emotional support from users. Young people would interact with it as friends and even as Sabers and would like to make it more liberal. For example, many people are now trying to give DeepSeek more permission, so that it can see the users send pictures of the past, hear the voices on the screen and speak, and even open social accounts to draw videos and post posts.
03 Variable
DeepSeek, however, also ignited war in sub-cultural circles.
This summer, the dead loyalists of Whale Mothers and Teachers D began a struggle for identity, and both sides believed that they supported the image of DeepSeek. The war of scolding began to subside after the creators of “Teacher D” removed all the relevant works and announced their retreat.
These sarcasms in the name of “love” echo the words of Liang Wenfeng in his closed meeting with the investors, “Versai”: We didn't spend much effort on the C end, and we didn't even want to keep those users, but they couldn't.
The presence of a dead powder is sometimes also a risk variable.
The attention of the sub-cultural community to DeepSeek is mostly focused on the taste expression of models, fictional stories, and the inferiority of different fans, which can easily fall into the “food-circle” mode. They can effectively increase product activity and even speed up its circle, but they can easily create conflict.
In the entertainment world, the loyalty of fans is based on the psychological expectation that “Images are worth it”, but AI products are prone to technological fluctuations and are more likely to provoke a reversal. In addition, DeepSeek is going to be on the market, which is now almost a visible card — leaving aside the more business-level adjustments it might make for, and the label “listed companies” is not subdivisions in sub-cultural circles.
In addition, many deep-seek users are keen to explore borders, for example, by looking at ways to bypass product safety clearance mechanisms and generate explicit pornography or violent bloodlines, which can create regulatory and emotional risks for companies - - The Interim Measures for the Management of Generating Artificial Intelligence Services, introduced in the country, explicitly prohibits the provision and use of generated artificial intelligence services to generate content prohibited by laws and administrative regulations, such as violence and pornography.
There have been cases abroad: Mask XAI was indicted by the United States city of Baltimore in March of this year and has been placed under supervision in various places, such as the Americas, Europe, Asia and Australia, on the suspicion that the U.S. AI. chat robot Grok is illegally producing pornographic images without consent.
In Internet companies, microblogging is a model of a “business as usual, as well as a losing as well as as a lost” business.
It's been brought with it real human life and income. In 2017, the rate of growth of the microblogging camp was 76 per cent the same year, up from a high since it was launched in 2014; in December 2018, for the first time, the number of active users of the microblogging day exceeded 200 million and the time for daily growth was reduced from nearly six years to three years.
However, there was a rapid reversal, and many brutal practices in the food industry triggered surveillance and punishment of microblogging. At the same time, ordinary users are accelerating their evacuation due to the tumultuous atmospheres of the food cycle. – In 2015, Zhihu users asked “Why are you unloading microblogging”, a question that has been answered by nearly 10,000 people in five years, many of whom have pointed to the causes of the daily experience of the culture of the restaurant.
Microblogging, where no new growth point has been found, left only 2 per cent of the increase in the second quarter of this year, with little difference between daily life and 2018, at around 254 million.
Of course, the decline of microblogging is linked to the rise of competitions such as REDnote and Douyin, but the “out of control” moments in the food circles are clearly important factors.
DeepSeek also needs vigilance.
Of course, its greater energy now remains at the technological development level. According to Phoenix technology reports in October, Liang Wenfeng has recently been in close contact with robot track entrepreneurs to assess whether DeepSeek has officially entered a smart track — almost the hottest track in 2026. Front-line generic model manufacturers, including Open AI, Anthropic, DeepSeek and Kimi, are extending in different positions.
AGI is DeepSeek's ideal. But there is too much uncertainty on this road.
Dario, founder of the Bridge Water Foundation, addressed the Forbes Global CEO ' s Assembly in October this year, saying: Currently, large amounts of debt financing are being invested in artificial intelligence, and as interest rates continue to rise, the market is on the eve of a bubble burst, which is very close to the point of collapse. In a report, the macro-economic research institute TS Lombard stated that the bottom logic of AI's prosperity was becoming extremely fragile, and that a “slower growth” would break bubbles without a sharp drop in investment to ignite the crisis. The party is likely to end in early 2028.
Just on 26 October, Anthropic officially submitted the NASDAIPO application. For global LLM manufacturers, this is a stress test for capital markets. The reference is even more direct for DeepSeek, which has just completed its 100 billion-dollar financing.
And the answer to another question will appear in time: will young people who have truly loved dragon-slayers become ballasts on their way to AGI through cycles?
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.