DeepSeek, a Chinese AI startup, has released a new large language model called R1. This model is exciting to scientists because it performs as well as OpenAI’s latest models but costs much less to use. R1 can solve problems in chemistry, math, and coding.
Expert @_KarenHao on what Deepseek may mean for the future of journalism and Gen AI.
"DeepSeek will be very inspirational for AI researchers who now realise they can develop approaches that require significantly less money."
✍️ A piece by @Marina__Adamihttps://t.co/yoUP9oYKZz
— Reuters Institute (@risj_oxford) February 3, 2025
It was released on January 20, 2025. The company has published the algorithm, which allows researchers to study and build on it. Using R1 costs about one-thirtieth of what it takes to run OpenAI’s models.
📝 Which is the best AI assistant?
The Turing’s Robert Blackwell helped @guardian try out some leading chatbots, asking #AI tools the same questions to understand their differences.
Read the full article: https://t.co/it1txXdpXE
— The Alan Turing Institute (@turinginst) February 3, 2025
DeepSeek has also made smaller versions of R1 for researchers with less computing power.
“While AI technology has provided hugely important tools… the business model depends on hype.”
Excellent piece on the significance of DeepSeek by @kenanmalik https://t.co/2YFyWRBr8E
— John Tasioulas (@JTasioulas) February 2, 2025
Mario Krenn, a researcher in Germany, said an experiment that cost over $300 with OpenAI’s model only cost under $10 with R1.
DeepSeek’s efficient and affordable R1
He said this big difference will affect how many people use it in the future. The fact that R1 was made in China is important.
“While privacy fears are justified, the main beef Silicon Valley has is that China’s chatbot is democratising the technology“
[DeepSeek has ripped away AI’s veil of mystique. That’s the real reason the tech bros fear it | Kenan Malik | The Guardian] https://t.co/gxZyqf80yC— Norio Nakatsuji (@norionakatsuji) February 2, 2025
It shows that China is catching up to the U.S. in AI development, even though computer chips used for AI are limited.
Large language models like R1 learn from huge amounts of text. They can predict the following parts of a sentence. But they can also make mistakes and have trouble reasoning through problems.
DeepSeek’s success highlights how important it is for AI research to be efficient and open. It could change the future of AI development.
