Did Deepseek really develop $ 6 million? No look

Deepseek, a Chinese -based artificial intelligence attempt, had a shock effect in the US technology sector at the weekend. The company announced that it has developed a large language model that can compete with giants such as OpenAI. Moreover, it is only in two months and 6 million ...
 Did Deepseek really develop $ 6 million? No look
READING NOW Did Deepseek really develop $ 6 million? No look
Deepseek, a Chinese -based artificial intelligence attempt, had a shock effect in the US technology sector at the weekend. The company announced that it has developed a large language model that can compete with giants such as OpenAI. Moreover, he claims that he has accomplished this with a budget of only two months and less than $ 6 million. So are these quite impressive claims real or is there a great exaggeration?

Deepseek announced that it created its new R1 model (based on the previous V3 model) using only a relatively few Nvidia H800 chips. The H800 chips are known as a product with a limited performance in line with the US export restrictions. Despite these limitations, the company says it has developed an artificial intelligence system that can compete with the best models of the West. Deepseek also stated that the cost of the R1 model is only 6 million dollars to strengthen its claims. This figure is only 3-5 %of OpenAI’s estimated costs for new generation models.

Is it too ambitious to be real?

One of the first reacting to Deepseek’s claims was SCALE AI CEO Alexandr Wang, one of the important names in the field of artificial intelligence. Wang claimed that he had reached the stronger Nvidia H100 chips by piercing US export restrictions for Deepseek’s model. Speaking to CNBC, Wang said, öyle As I understand, Deepseek has about 50,000 H100. They cannot clearly express the existence of these chips, because this is contrary to export control. ”

Elon Musk also interpreted Wang’s statements through the social media platform X and said “absolutely”. US technology analyst Ted Mortonson said that Deepseek’s claim that he might have developed a model at this level with H800 chips.

Gavin Baker, the head of the investment company Atreides Management, argued that Deepseek’s cost claims were misleading. Baker, Deepseek announced that the cost of $ 6 million, did not cover previous research and trials, he said. Baker, Deepseek’s technical document, this figure does not contain the costs of previous research and ablation experiments on architectures, algorithms and data.

Experts speaking to Reuters also make a similar discourse. In the research article, this cost refers to the use of the chip especially in the latest educational study and did not refer to all development costs. In general, as accepted in the AI ​​world, educational work is the visible part of the iceberg in terms of total cost. It is a fact that more chip is used in previous stages of artificial intelligence.

Is the cost very important?

We cannot know exactly the costs of these models without explaining exactly how much Deepseek, R1 and other models spend in the process of training and development. However, it is necessary to ask; Are the costs so important?

Obviously, of course, costs are very important for any product, but in China, it should not be evaluated by closing the eyes. As it is known, the US has limited China’s access to advanced equipment, chips and semiconductor technologies. In this way, the US wants to limit China’s growth in artificial intelligence.

Deepseek’s V3 and R1 models show that China somehow can openly compete with the models of OpenAI, the most advanced artificial intelligence company in the USA, and can put forward models in some areas. So let’s ask again, do you think the costs are so important? Or does it matter how to reach this level?

Here, almost all of the experts we include explanations about costs say that the progress of Deepseek is admirable. Deepseek’s most praise is that these models have launched these models as open source. This is a great service for the world of artificial intelligence.

Comments
Leave a Comment