DEPSIC: Everything you need to know about the AI ​​Chatb OTT app

Dippic has gone viral.

Chinese AI Lab Deepske entered the mainstream consciousness later this week Its chatb ot t t app reached the top of Apple Pal App Store Charts. Dippic’s AI models, which were trained using computing-efficient techniques, Vall L Street has led analysts – And technologies – U.S. U.S. in the race It can maintain its lead and question whether AI will sustain the demand for chips.

But where did the dippic come from, and how did it grow so quickly in international fame?

Dippic’s merchant origin

Dippic has been endorsed by High-Flyer Capital Management, Chinese Quantitative Hedge Fund, which uses AI to report its trade decisions.

AI’s enthusiastic Liang Wenfeng co-founded a high-flyer in 2015. Vanfeng, a student at Zejiang University, began dubling in trading, launched a high-floor capital management as a hedge fund in 2019, focusing on the development and deployment of AI algorithms.

In 2023, the High-Flyer AI Tools began a dippick as a dedicated lab for a separate research from its financial profession. With a hi-flyer as one of its investors, the lab enters its own company, also called dippic.

From the first day, Deepske created its own datacenter clusters for model training. But like other AI companies in China, Deepsin on hardware to the U.S. Has been affected by the export restriction. To train one of its recent models, the company was forced to use NVIDIA H800 chips, a less powerful version of the chip available to US companies.

The technical team of Deepsik has asked Young to skip. Company Reportedly aggressively recruited Doctorate AI Researchers from Top Chinese Universities. Dippecic any computer viz .N hire people without a background According to the New York Times to help its technology better understand a wide range of different topics.

The stronger models of the dippecic

Deepcke unveiled the first models of Deepsk Coder, Deepsk LLM and Deepsic Chat-its first models in November 2023. But it was not until last spring, when the startup published the family of its upcoming-Jan-Jan Deepsk-V2 models, that the AI ​​industry began to take note.

Deepsik-V2, a common-intended text- and image-annalizing system, performed well in various AI benchmarks-and it was much cheaper to run than comparable models at that time. It was forced to reduce the consumption prices of some models, including Deepsik’s domestic competition, including Bidens and Alibaba.

Director-V3Launched in December 2024, only added to the infamous Deepsk.

According to the Internal Benchmark test of Deepsik, both downloadable, openly available models, such as Dippacic V3 Meta Food post And “closed” models that can only be successful by API like OpenAI GPT.

Dippic’s R1 “Logic” model is equally impressive. Dippecic claims that published in January R1 Key exhibits the O1 model of OpenAI at the benchmark.

Being a logic model, R1 effectively examines itself, which usually helps to avoid some of the difficulties of traveling models. To reach solutions compared to the typical non-resining model-usually from seconds to seconds long-turning models take a little longer. The sidewalk is that they are more reliable in domains like physics, viz .an and mathematics.

However, there are losses to other models of R1, Dippic V3 and Deepsik. Being a Chinese-developed AI, they are subject to Fore By China’s Internet Regulator to ensure “the main socialist values ​​embody the main socialist values.” In the Dippick’s Chatb OTT application, for example, R1 Tiananman will not answer questions about Square or Taiwan’s autonomy.

Disturbing approach

If Deepsik has a business model, it is not clear what the model is, exactly. The company offers its products and services well at market value – and offers others for free.

The way Deepsik says, efficiency progress has enabled it to maintain the competitiveness of the extreme cost. Some experts Controversy However, the figures provided by the company.

Whatever the case may be, developers have taken the models of the dippick to Delo, which is not an open source because this sentence is generally understood but available under the permit license that allows for commercial use. According to Clem Dallang, CEO of Hugging Face, one of the platforms hosting models of Deepsk, Developers on the embrace have created more than 500 “derivative” models of R1 Which has jointly raised 2.5 million downloads.

Deepsik’s success has been against big and more established rivals Described as “Upping AI” And “The new era of AI Brinkmanship.” The success of the company was at least responsible Nvidia’s stock price fell 18% on MondayAnd for Obtain a public response OpenAI CEO from Sam Altman.

What may be the future of Dippic, it is not clear. Improved models are given. But the US The government appears Careful with what he believes as a harmful foreign influence.