Speaker 1: It's almost Chinese New Year. Before the holiday, the domestic AI model manufacturers released a lot of Chinese New Year gifts. China's big model black horse DeepSeek, Kimi, Yinggang OpenAI, said it was a warning spray, but it was not an accidental outbreak, but a power that has lasted for a long time. Google's former CEO said bluntly that the Chinese are too smart. I underestimated them. There are several projects in the field of AI that have made it to the United States.

Speaker 2: Chinese programs, one for example is called DeepSeek, looks like they've caught up and this is even after these extraordinary efforts by the Trump and Biden administration to withhold high-powered chips. The Chinese are clever and they understand the power of a new kind of intelligence for their industrial might."

Speaker 1: It can be said that the domestic AI model will be able to redefine the world coordinates of AI development in the early days of the Spring Festival. Foreign bloggers are already anxious, saying that DeepSeek may kill OpenAI and ChatGPT.

Speaker 3: This Chinese company could kill OpenAI and ChatGPT because they just created an AI model that is better than ChatGPT for absolute head-scratcher.

Speaker 1: I heard that the current AI model can also realize the vision of the future. On January 20, DeepSeek, a Chinese AI company, released the DeepSeek R1 model. With its unique reinforcement learning training method AI showed the moment of awakening of human beings for the first time When solving complex problems the model will suddenly stop and think and correct the reasoning path independently and even mark it with a resource language and so on This is impossible to be mistaken This is too awesome Artificial intelligence experts believe that in math, programming and reasoning and other tasks it has reached a level equivalent to OpenAI OE At the same time, the application program programming interface the cost of calling has been reduced by 90% to 95% More importantly the experimental version of this model DeepSeq RE0 Today we have R1, so DeepSeek is the company behind all of this, DeepSeek R1 is the actual

Speaker 4: model. We've talked about it before, but today it's unleashed, if you will, it's open source. This post is a few hours old, it's got 2 million plus views, and as it says, DeepSeek R1 is here. Performance on par with OpenAI 01, kind of that big model, the reasoning model that we have access to. It's a fully open source model. Notice it's as good as the 01 model across most of the common tasks, including the AIME 2024, the sort of the high-end math problems. As a bonus, they actually have the open-source distilled models.

Speaker 1: It doesn't have to be a mysterious name such as ASI or Strawberry Plan Sometimes it can also have a profound impact on the original algorithm and learning curve AI teacher Shindun was asked by the media how to view Chinese AI He said it may take some time But in the end, China will overtake by strength

Speaker 5: Something that many AI researchers didn't predict was that China would catch up with the West in terms of AI development So how do you feel about that and what are the consequences?

Speaker 6: I don't think they're quite caught up yet They're very close, though. America's going to slow them down a bit by trying to prevent them having the latest NVIDIA chips. NVIDIA maybe can find workarounds. And what that's going to do, if the embargo is effective, it's just going to cause the Chinese to develop their own technology. And they'll be a few years behind, but they'll catch up. They've got better STEM education than the U.S., so they've got more people who are better educated. I think they're going to catch up.

Speaker 4: The Chinese DeepSeek R1 model is out and it's fully open source. It beats or matches O1, the best available model from OpenAI. It's capable of creating other models through distillation – we'll get to what that means in a second. But it's able to create incredibly powerful tiny models that are as strong as, for example, the GPT-4-O model. And all of this is in of itself is kind of earth-shattering. Keep in mind, this is open source. You can run it on your home computer. You can use it for your business. You can use it to create your own models. This thing has just been let loose into the wild.

Speaker 1: The high cost-effectiveness and open-source behavior are shocking, calling for the primordial spirit No wonder foreign bloggers say that overseas business people have a strong reaction to the release of DeepSeek R1

Speaker 7: Strongly to the DeepSeek R1 release It is an incredible model, completely open-source, open weights They shared all of their secrets of how it actually works and how they were able to accomplish it A flood of open-source thinking models is right around the corner

Speaker 1: Interestingly, within two hours after the release of DeepSeek R1 Another domestic manufacturer, Kimi, also launched a new model K1.5 that represents OpenAI OE in terms of mathematical code and multi-modal reasoning ability According to their official introduction, from the standard test results K1.5 multi-modal thinking model achieves SOTA-level multi-modal reasoning and universal reasoning ability The new model fully represents OpenAI's full-blown version of OE in terms of mathematical code and multi-modal reasoning ability This is also the first multi-modal reasoning performance of a company other than OpenAI to reach OE's official version According to media reports, Kimi explored a base model that is different from DeepSeek V3 through the combination of long-term models and short-term models and the specialization RL of short-term models to maximize the ability to classify OE models It can be said that the domestic large-scale model team represented by DeepSeek attempts to compare the most advanced large-scale models in terms of performance and shows the ultimate cost-effectiveness in pricing Even Jack Borg said that the gap between China and the United States is extremely narrow If you don't work hard, it's over Call on the United States to establish AI standards ahead of time

Speaker 8: There's this great Chinese model that just came out This company DeepSeek They're doing really good work It's a very advanced model I just think like, we should all want the American companies to win at this, right? It's like, this is like a huge geo political competition and like China's running at it super hard. And like, we should want the American companies in the American standard to win. And like, if there's going to be an open source model that everyone uses, like we should want it to be an American model, but like every step along the way, if the government is sort of making that harder rather than easier, then that's, I don't know. I mean, there's, there's an extent to which, okay, the American tech industry is leading. So maybe the government can like get in the way a little bit and maybe the American industry will still lead, but I don't know. It's, I think it's getting really competitive. And I think like it's easy for the government to take for granted that the U S will lead on all these things when I think it's a very close competition and we need the help, not, We need them to not be a force that's helping us to do these things

Speaker 1: In addition to what I just mentioned, other domestic large models also performed well On January 15th, Minimax released a new open source model series, Minimax01 In multiple standard and internal benchmark tests, the performance of Minimax01 is equivalent to the top models of GPT-4O, Cloud 3.5, and SONNET Especially in terms of long-term and long-term processing, it is obvious that now the domestic and foreign large model manufacturers are chasing me In a short period of time, the demand for computing power from large model manufacturers continues to increase. Behind the OpenAI GPT series, domestic large-scale models are closing in. Let's wait and see what happens in the future. I boldly predict that domestic large-scale models will continue to shine on the world AI stage. Overcoming is not the purpose, leading is.