Suddenly, DeepSeek is everywhere.
Its R1 model is an open source and is said to be trained for some costs of other AI models.
This fatal combination questions how much money you need to attack Wall Street violently, fall down high -tech stock, and develop an excellent AI model for investors. DeepSeek engineers argue that R1 has been trained for 2,788 GPUs.
Deepseek’s cost efficiency is also the idea that larger models and more data can lead to improvements in performance. It may be difficult to understand what is happening in the enthusiastic conversation of DeepSeek’s abilities, the threat to AI companies like Openai, and what is happening in the surprising investor threat. However, AI experts with veteran experience have become heavier from a valuable perspective.
DeepSeek proves what AI experts have said for years: Bigger is not good
Deepseek, based in China, was hindered by trade restrictions and access to the NVIDIA GPU, and had to be creative in the development and training of R1. It was a revelation to investors that they were able to achieve this feat in just $ 6 million (this is not a lot of money under the conditions of AI).
But the AI expert was not surprised. “We asked why we had to stick to building the biggest model in Google. Why are you going to the size? What functions are you trying to achieve? Why you are the maximum. Why did you get upset about not having a model, and posted a famous Timnit Gebru from Google by calling for AI bias?
Mashable light speed
The tweet may have been deleted
Face’s climate and AI lead Sasha Luccioni pointed out that AI investment is unstable for marketing and hype. “It is wild that the single (high -performance) LLM is enough to achieve this, and the single (high -performance) LLM can achieve its performance,” said Luccioni. I mentioned it.
The tweet may have been deleted
Clarify why Deepseek R1 is so big
Deepseek R1 has been relatively functioning on the Openai O1 model on the key benchmark. Mathematics, coding, and general knowledge tests have slightly exceeded O1, equally or down. In other words, there are other models that average users can do in the same way, such as the mankind’s Claude, Google Gemini, and META’s open source model LLAMA.
However, R1 causes such a frenzy because it is not expensive to make. “I’m not smarter than the previous model. I trained cheaply,” said Gary Marcus, a researcher in AI.
The tweet may have been deleted
The fact that DeepSeek has been able to build a model that competes with Openai model is very noteworthy. Openai posted in X Andrej Karpathy “Does this mean that Frontier LLMS does not require a big GPU cluster? No. No. demonstration.”
The tweet may have been deleted
Professor Warton AI said, Ethan Morrick, not the ability, but a model that people are currently accessing. “Deepseek is a really good model, but it’s not a good model in O1 or Claude.” “However, it is free and attracting much attention, so many people who used the free” mini “model have been able to do early inference AI in 2025. I think there is. “
The tweet may have been deleted
Get one of the open source AI models
Deepseek R1 Breakout is a great victory for open source supporters who claim to guarantee democratization, transparency, innovation, and health competition to provide democratization of powerful AI models. For those who think that China is more than the United States with AI, the right idea is that “open source models are out of the closed models”, said Yann Lecun, a meta chief AI scientist. 。
The tweet may have been deleted
AI’s expert Andrew NG, a computer scientist, did not clearly mention the importance that R1 is an open source model, but how DeepSeek’s confusion is how to benefit developers. I emphasized.
“Today’s” Deepseek Selloff “-Deepseek V3/R1 destroys the high -techeco system is another sign of the application layer,” said NG. I am. “Hyper competitive basic models are perfect for those who build applications.”
The tweet may have been deleted
Topic Artificial Intelligence Deepseek
