본문 바로가기

U.S stocks [2025] ISSUE arrangemet

Maybe DeepSeek is spreading

반응형

Maybe DeepSeek is spreading a lot of misconceptions in the US and Korea with unfounded content...

A clear and concise article has been posted to correct this.

I think it would be good to refer to it.

In fact, DeepSeek didn't just drop from the sky all of a sudden, but the Generative AI community has been announcing enormous results in the last nine months.

1. Is DeepSeek a questionable company that has suddenly emerged?
-> Not true. DeepSeek has been continuously developing AI models since November 2023, and they were well known to AI researchers.

2. DeepSeek's Model Development Cost Can't Be $6M
-> False Claims. The training cost analysis published by DeepSeek is largely consistent with industry standards, and additional R&D costs are the same for all AI labs.

3. If DeepSeek made it too cheap, did American AI companies waste money?
-> Wrong. DeepSeek has maximized efficiency, but the scaling law that more computing resources increase AI performance is still in effect. American AI companies will still continue to invest in GPUs, and it does not directly affect NVIDIA's GPU market dominance.

4. DeepSeek is just a copy of the technology of American AI companies?
-> Not true. DeepSeek has achieved its own technological innovation, typically using more efficient memory compared to existing Transformer models such as Multi-lattent Attention, and increasing efficiency in the process of optimizing Reinforcement Learning (RL) and training multiple GPUs. On the other hand, DeepSeek shares its technology as an open source, while OpenAI and Anthropic operate the model in a closed manner. It should be praised.

5. DeepSeek trained the model using ChatGPT data
-> No basis. OpenAI claimed that DeepSeek took data from ChatGPT, but there is no concrete evidence. If DeepSeek directly generated and learned ChatGPT data, this is a violation of OpenAI's policy, but it doesn't matter if it used a public dataset. DeepSeek's success is not due to ChatGPT data, but to advances in its own techniques and learning techniques.

the beginning of a life

320x100

'U.S stocks [2025] ISSUE arrangemet' 카테고리의 다른 글

Adam Smith said in classical  (5) 2025.02.06
Basic elements of ICT trading  (8) 2025.02.05
Prices rarely move in a straight  (5) 2025.02.05
Market Structure Shift  (6) 2025.02.05
Liquidity - Key to ICT Trading  (6) 2025.02.05