Lao Gao talks about Deepseek: absolutely no plagiarism of ChatGPT, bypassing the NVIDIA CUDA platform through underlying technology

The well-known YouTuber Lao Gao released a video yesterday to talk about his views on Chinese AI startup DeepSeek, saying that DeepSeek's alleged plagiarism of ChatGPT is obviously not true, and that DeepSeek's success lies in bypassing Huida's CUDA computing platform, but experts dispute this claim. (Synopsis: New US AI Anti-China Act: Using DeepSeek is a felony, up to 20 years in prison and a fine of 1 million magnesium) (Background supplement: Brief analysis" Why can DeepSeek affect the encryption money market? Chinese AI startup DeepSeek recently released two large models, "DeepSeek-V3" and "DeepSeek-R1", which shook Silicon Valley because of their low cost and comparable performance to OpenAI, and were considered to be likely to rewrite the rules of the game in the field of AI big models. Lao Gao talks about the DeepSeek impact In this regard, YouTuber Lao Gao released the video "DeepSeek, which caused a global shock, has ignited a war without gunsmoke between China and the United States", talking about his views on DeepSeek. Lao Gao pointed out that DeepSeek's success lies not only in the performance of its AI models comparable to ChatGPT, but also in its extremely low development costs, DeepSeek spent only $5.6 million to develop its top models (the actual cost is currently widely debated, probably not so low), about one percent of OpenAI, which has greatly improved DeepSeek's competitiveness in the field of AI and triggered a re-evaluation of the cost and efficiency of AI development. Lao Gao believes that DeepSeek's biggest breakthrough lies in its Open Source strategy, and unlike OpenAI's closed-source model, DeepSeek makes its AI models public, allowing anyone to download and run them locally, even for commercial use. This move not only significantly lifts the threshold for enterprises and individuals to use AI, but also has a huge impact on companies such as OpenAI that rely on closed-source models to make money. DeepSeek plagiarized ChatGPT? In addition, Lao Gao said that the claim that DeepSeek was accused of plagiarizing ChatGPT is obviously not valid, because ChatGPT adopts a closed-source model, and its internal operations cannot be copied externally, and DeepSeek, as an Open Source project, all code and data are open and transparent, and OpenAI can already detect plagiarism. It's like some well-known operating systems are closed source and not easy to imitate. Mr. Gao described DeepSeek's work as similar to ChatGPT only in inspiration, rather than a direct copy, because no one could see each other's "cooking secrets." Lao Gao also claimed that DeepSeek's success lies in its ability to bypass Huida's CUDA computing platform, which has always been like a nuclear power plant, converting powerful computing chip capabilities into stable and efficient computing resources, forming its insurmountable technical barriers. CUDA is a parallel computing platform and programming model provided by Huida that allows developers to utilize Huida GPUs for high-performance computing. According to Lao Gao, DeepSeek can bypass hardware restrictions on training speed and no longer rely on CUDA, which means that no matter which company's chip is used, as long as it can access this technology, it has the opportunity to obtain maximum computing performance without using CUDA, further shaking Huida's monopoly position in the computing market. Does DeepSeek really bypass the Huida CUDA framework? However, in response to Lao Gao's statement, some experts said to the moving area that the argument in Lao Gao's film is wrong, and the current Deepseek is definitely using Huida graphics cards for computing, and still uses the CUDA platform, which is by no means "bypassing CUDA" as Lao Gao claims. Experts pointed out that Lao Gao may have misunderstood the recent news, Deepseek has recently been rumored to be preparing for a possible future block, even if it cannot use Huida graphics cards, it also intends to use Chinese domestic GPUs as a source of computing (it is also said that China still obtains a large number of Huida chips through gray means), but this does not mean that the company's current model is "bypassing CUDA" in operation. Previously, Tom's Hardware had reported that when DeepSeek used the Huida H800 chip for training, some functions used the Huida underlying hardware instruction PTX language, rather than the high-level programming language CUDA, Huang Lei, associate professor at Beihang University, analyzed that bypassing CUDA means that DeepSeek can directly do some new development based on the GPU's driver functions to achieve more nuanced operations. Related reports OpenAI shrunk! Altman: No plans to sue DeepSeek, focus on building better products Forced by DeepSeek? OpenAI launches new inference model o3-mini for free! Altman admitted for the first time: the closed-source strategy is wrong, DeepSeek raises financial security concerns, and Taiwan's public officials ban it! Legislator: We can't just ban, but also invest in the development of AI "Old talk about Deepseek: absolutely did not copy ChatGPT, bypassing the Huida Cuda platform through the underlying technology" This article was first published in BlockTempo "Dynamic Trend - The Most Influential Block Chain News Media".