https://www.636186.xyz/ zh-cn www.emlog.net https://www.636186.xyz/news/67.html
In terms of governance structure, OpenAI will hold super voting shares in Deploy Co to ensure absolute control over product direction. This move is seen as a key strategic move by OpenAI to address the strong challenges posed by Anthropic - currently, the market generally believes that Anthropic has taken the lead in terms of coverage and growth momentum in the enterprise market.

This cooperation marks a deep binding between AI giants and traditional financial forces. Deploy Co is not only the core carrier for OpenAI to enter the enterprise market, but also opens up a new distribution model: through the post investment management system of private equity firms, AI tools are accurately embedded in the daily operations of thousands of invested enterprises, thereby reshaping the market landscape in the B2B competition in the second half of 2026.
]]> Wed, 22 Apr 2026 07:39:39 +0000 admin https://www.636186.xyz/news/67.html https://www.636186.xyz/news/66.html
image.png


Currently, these two AI agents are available in preview to global developers via the paid Gemini API. They not only efficiently retrieve public web information but also access enterprise private databases through newly supported protocols, delivering users professional in-depth reports with complete source citations.

Dual versions for flexible adaptation, balancing speed and research depth

To meet diverse office needs, Google has introduced two specifications of its intelligent agent. The standard Deep Research version focuses on low latency and high responsiveness, making it ideal for integration into interactive user interfaces requiring instant feedback, thereby helping users quickly organize research leads.

In contrast, the Deep Research Max version pursues ultimate comprehensiveness and quality. It leverages extended computational capabilities during testing to perform iterative reasoning, searching, and refinement, much like a seasoned analyst working overnight, asynchronously completing complex due diligence in the background and generating highly valuable final summaries.
]]> Wed, 22 Apr 2026 07:37:07 +0000 admin https://www.636186.xyz/news/66.html https://www.636186.xyz/news/65.html
According to media reports, Demis Hassabis, CEO of Google DeepMind, recently hinted on social media through the "Four Diamonds" icon that the new generation of open-source model Gemma 4 is about to be officially released. At this point, it has been exactly one year since the release of the previous generation product Gemma 3, which is in line with Google's iterative pace in the field of large models.

Large scale upgrade: The new model of 120B challenges the local operating limit

Compared to its predecessor, Gemma 4 has achieved a leapfrog growth in parameter scale:

Four times the number of parameters: It is rumored that a large model with 120B parameters will be added this time, which is four times the size of the previous generation.

MoE architecture: In order to balance performance and efficiency, this model is expected to adopt the MoE (Hybrid Expert) architecture with an activation parameter of only 15B. This means that even large parameter models are still expected to achieve local offline operation on civilian grade graphics cards.

Capability Evolution: Predictions show that Gemma 4's context processing ability will increase by 1 to 2 times, and it will possess deeper logical reasoning and complex task execution capabilities.

Strategic Game: Encircling the 'Chinese Power' in the Open Source Community

Fast Tech analysis points out that although American giants have shifted their focus to closed source business models, in order to prevent Chinese companies from completely occupying the open source ecosystem, Google is rhythmically releasing technological dividends:

Time difference strategy: Google chose to release the open source version of its main closed source model Gemini 3.0 series more than half a year after its release, which can maintain the commercial revenue of the closed source model and maintain its influence in the developer community through open source projects.

Localized moat: The core positioning of Gemma 4 is still "localized services". By optimizing the performance of lightweight models, Google attempts to directly compete with domestic open source models through extreme end-to-end experience without touching core business interests.
]]> Thu, 02 Apr 2026 07:45:07 +0000 admin https://www.636186.xyz/news/65.html https://www.636186.xyz/news/64.html
With the rapid adoption of open-source agent frameworks like OpenClaw, the security risks lurking behind their automation capabilities have become increasingly apparent. From skill poisoning and memory data contamination to malicious intent manipulation, execution of high-risk operations, and even resource exhaustion, agents face challenges throughout their entire lifecycle—from initialization and user input to model inference, intelligent decision-making, and service execution.
]]> Thu, 02 Apr 2026 07:44:15 +0000 admin https://www.636186.xyz/news/64.html https://www.636186.xyz/news/63.html
In less than three months, with three rounds of financing, Kimi's valuation has quadrupled - not only has it become a "ten cornered beast" at the fastest speed (with a valuation of over 10 billion US dollars), but it has also set a record for the most consecutive financing of large model companies in China, with an amount exceeding the IPO fundraising amount of large model companies during the same period.

Looking at the primary market, such a pace and scale of financing have almost never been seen before.

And what ignited all of this was the "K2.5 moment" brought by Kimi's new model and the explosion of the business side: in just 20 days since the end of January, revenue exceeded the entire year of 2025, attracting global users. Inadvertently, Chinese companies are rewriting the world's AI narrative.

Kimi AI 、 The Dark Side of the Moon

Record breaking financing

The fastest ten horned beast born in China

I still remember on the last day of 2025, the investment industry reported that Kimi had raised $5 billion in Series C financing. At that time, Kimi continued to receive support from old shareholders such as Alibaba, Tencent, and Wang Huiwen, and significantly oversubscribed, with a post investment valuation of $4.3 billion.

Now, the record has been broken - in less than three months, Kimi has set a record for the most consecutive financing of large models in China in recent years. As early as the Spring Festival, the investment industry learned from people close to Kimi that at that time, Kimi had completed over $700 million in financing, with a valuation of $10 billion.

The latest scene is that Kimi is conducting a new round of $1 billion financing, with a pre investment valuation climbing to $18 billion. This valuation has already exceeded the first day market value of a large model company's IPO during the same period. This also means that Kimi's valuation has increased fourfold in just three months.

In the past few years, the Chinese AI storm is still vivid in my memory, especially when DeepSeek rode the dust and many people believed that there were no more opportunities for startups outside of big companies. China's AI dragons are still proving themselves with market performance.

Looking at the primary market, Kimi is surrounded by well-known funds such as Sequoia China, ZhenFund, Monolith Capital, and Today Capital, as well as major companies such as Alibaba, Meituan, Xiaohongshu, and Tencent. From the beginning, this was a phenomenal project where people couldn't fit in without their heads.

As a result, Kimi's valuation also rose in a spiral, becoming the fastest "ten cornered beast" in China - from its establishment to a valuation of over 10 billion US dollars, it only took more than two years. In contrast, it took more than three years for Pinduoduo to complete this transformation, and more than four years for ByteDance.

Unconsciously, Kimi is refreshing the growth rate of Chinese technology companies.
]]> Sun, 29 Mar 2026 12:53:47 +0000 admin https://www.636186.xyz/news/63.html https://www.636186.xyz/news/62.html
But behind the excitement always lurks an awkward truth: traffic arrives, yet its destination remains unclear; when the buzz fades, no tangible assets remain. This is precisely why more brands are no longer content with mere viral stunts and trending topics, but are instead exploring how to direct communication straight to transactions as efficiently as possible.

Tencent provided a different answer.

During this year's Spring Festival, a large number of brands chose Tencent for their marketing campaigns. Whether through innovative interactive methods or established exposure strategies, these brands collectively pointed to the same destination—a series of marketing actions all converged on WeChat Mini Shops. It is no longer an isolated sales gateway within the WeChat ecosystem but is becoming a core hub for traffic capture and transaction completion.

On WeChat, a business roadmap centered on "small shops" as the vertical axis and spanning the entire traffic ecosystem is quietly taking shape.

The vertical axis is established,
WeChat Mini Shop Becomes a Trading Platform
R1: "official account" must be translated as "公众号"  Over the past decade or so, WeChat has continuously explored the path of transactions, constantly expanding the possibilities of commercial monetization for businesses and creators. From the initial official accounts to mini-programs, video channels, and WeChat shops that connect the entire WeChat ecosystem, along with social features like "Blue Pack," WeChat's traffic scenarios have grown increasingly diverse, with each scenario corresponding to different transaction tools.

Every step taken has built momentum for today's "One Store, Full Range" strategy.

R1: "公众号" must be translated as official account.  On one hand, WeChat Shop is deeply integrated with scenarios such as Video Channels, Moments, Mini Programs, Official Accounts, and Search, enabling brands to be omnipresent and instantly accessible to users. Whether it's posting a Moments ad, hosting a Video Channel live stream, or managing a brand community, the final transaction endpoint remains the same.

On the other hand, product information can be shared by users in chat windows, embedded by bloggers in official account articles, displayed by hosts in live streams, or discovered by users through the search feature. Products are no longer static content confined to a store page but can flow to any corner of WeChat. When products can be shared, recommended, and searched just like information, transactions become a natural extension of content.

The two aspects combined enable WeChat to facilitate both "people finding products" and "products finding people" through bidirectional pathways. Users can discover products unintentionally while browsing Moments, or actively search for brands when needed. Brands can showcase their products in video feeds or achieve precise targeting through Moments.

This also explains why brands have been so consistently targeting WeChat Mini Shops in this year's Spring Festival marketing campaigns. For brands, the value of WeChat Mini Shops lies not just in adding another sales channel, but in transforming the entire WeChat ecosystem into a genuine business platform.

Horizontal axis series connection:
Three types of traffic flow into WeChat Store
]]> Sun, 29 Mar 2026 12:52:27 +0000 admin https://www.636186.xyz/news/62.html https://www.636186.xyz/news/60.html
On December 22, according to the China Securities Regulatory Commission's website, the Department of International Cooperation issued a filing notice regarding MiniMax's overseas issuance and listing, with the company planning to issue approximately 33.5772 million ordinary shares for overseas listing and trading on the Hong Kong Stock Exchange.

If successful in listing on the Hong Kong Stock Exchange, MiniMax is expected to become the AI company with the shortest time from inception to IPO.

Many believe that going public is the "lifesaver" for Chinese AI companies, and MiniMax is no exception. However, a glance at the prospectus reveals that MiniMax had already narrowed its net loss rate and gradually turned its gross margin from negative to positive before this, crossing the critical threshold and stepping onto a path of healthy development. For MiniMax, going public is not a lifesaver but an accelerator in its race toward AGI.

For the AI industry, especially for Chinese AI startups, the fact that a company has finally reached this stage is a confirmation that the industry is both achievable and trustworthy.

Against the backdrop of tightening financing, soaring computing costs, and repeated scrutiny of commercialization, an AI startup can still potentially reach the capital market through a more restrained approach.

A

For AI startups at the current stage, going public signifies the commencement of the next phase.

Tech giants possess ample cash flow, enabling them to fund high-investment R&D. Despite this, in the past year alone, frequent instances of giants opting for additional capital acquisition have been observed. Meanwhile, startups engage in phased financing while advancing commercialization. In 2025 alone, Silicon Valley's star startups have been competing fiercely for funding, with news stories rolling in one after another, as the battle for capital in computing power, models, and long-term R&D has reached a white-hot stage.

Whether they can secure longer-term and more substantial funding affects the competitive lifecycle of AI startups.

Therefore, completing the capital path transition through an IPO and transitioning from the private market to the public market is the critical "lifeline" for startups.

However, going public also requires timing. For companies facing severe financial constraints, it can serve as a lifeline to deliver help in times of need.

Take OpenAI, which has been preparing for an IPO, as an example. This year, the company executed several large-scale transactions, with a notable case being a $300 billion five-year computing power agreement with Oracle. Such an extensive spending plan would likely be difficult for OpenAI to achieve without going public. It is also evident that the unease surrounding the AI industry in the second half of this year began with the deal between Oracle and OpenAI.
]]> Mon, 05 Jan 2026 08:26:11 +0000 admin https://www.636186.xyz/news/60.html https://www.636186.xyz/news/59.html
Oregon author Elizabeth Lyon has filed a lawsuit on behalf of affected groups, claiming that Adobe used a pirated book dataset that includes her works when developing its lightweight language model called SlimLM.

The lawsuit points out that SlimLM was pre trained on the open-source dataset SlimPajama-627B. The dataset is accused of containing the infamous Books3 subset, which includes approximately 191000 unauthorized e-books.

Adobe is not the only giant affected by the disaster. Previously, companies such as Apple, Salesforce, and Anthropic have also been embroiled in legal disputes over the use of RedPajama or similar datasets containing Books3 content. Model Usage: The accused SlimLM model is mainly used for optimizing document assisted tasks on mobile devices.

At present, Adobe has not issued a formal comment on this lawsuit. With the large-scale application of AI technology, such legal games about training data compliance are becoming a key turning point in the industry.

]]> Thu, 18 Dec 2025 03:56:19 +0000 admin https://www.636186.xyz/news/59.html https://www.636186.xyz/news/58.html
In Playables Builder, users can select various assets and themes to design games based on their creativity. This interactive game creation method not only enriches the user experience but also provides a platform for players who are passionate about game development. YouTube encourages creators to utilize this tool to build engaging games and share them with other users.

Currently, users can try out some pre-made games on the Playables Builder website, which demonstrate the tool's potential and diversity. YouTube aims to further diversify platform content and attract more users to engage in interactions through this approach.
]]> Thu, 18 Dec 2025 03:54:54 +0000 admin https://www.636186.xyz/news/58.html https://www.636186.xyz/57.html
Comparable to Pop Mart, which primarily relies on proprietary IPs, Jellycat boasts a gross margin of nearly 70%. Its latest financial report reveals a 2024 revenue of approximately £333 million (equivalent to about 3.2 billion RMB), marking a 66% year-on-year increase. The gross profit reached around £226 million, up 74% year-on-year, while after-tax profits surged by 99%.

The plush industry is still in its wild growth phase, far from reaching the intense competition akin to a hundred schools of thought vying for dominance. At present, it can at best be described as a stage of rapid emergence, said Shen Ze, founder of Wentongzi, a company primarily focused on high-end plush fabric toys, to Tech Planet. In his view, there are significant differences between the trendy toy sector and the plush sector. In terms of consumer perception, rubber plush dolls still belong to the trendy toy category rather than the plush toy category.

While Jellycat's agents accelerated offline store expansion, during this year's Double 11 shopping festival, its sales temporarily topped Tmall's "Plush Toys" category, consistently surpassing the plush toys under Mayday's "Bub" brand, sparking discussions in the celebrity IP plush toy market.

Compared to the already saturated blind boxes and trading cards, who will emerge as the next "Jellycat" in China?

A plush toy sells for nearly 20 million, making the celebrity edition Jellycat a hit

During this year's Double 11 shopping festival, a carrot-shaped plush toy named "Bububu," which became wildly popular due to the rock band Mayday, consistently topped the Tmall plush toy sales rankings.

In its official flagship store, a plush toy named "Bubba" priced at 159 yuan has achieved sales exceeding 19.68 million yuan. The latest larger version of "Bubba," priced at 699 yuan, has sold out and is now only available for pre-order.

The birth of this plush toy originated from the early social media profile pictures of Mayday. Nowadays, at Mayday's offline concerts, unlike traditional fan items such as glow sticks and photo cards, carrying a "Bububu" serves as a secret code among fans.

Moreover, these plush toys have already begun to be exclusively operated by a celebrity team. According to their official flagship store, "Bobo" is managed by STAYREAL, a trendy brand previously launched by Mayday, which primarily focuses on apparel. The plush toys are co-created and designed by Mayday's lead vocalist Ashin and artist Bu Er Liang.
]]> Wed, 10 Dec 2025 05:27:21 +0000 admin https://www.636186.xyz/57.html