DeepSeek R1 reportedly thinks and reasons like OpenAI o1
Nov. 21, 2024.
2 mins. read.
8 Interactions
A new Chinese AI considers each question carefully, trying to avoid mistakes by essentially fact-checking itself.
A Chinese Artificial Intelligence (AI) lab named DeepSeek has introduced a large language model (LLM) called DeepSeek-R1, TechCrunch reports. According to DeepSeek, DeepSeek-R1 can think and reason much like OpenAI’s o1 model.
Reasoning means that DeepSeek-R1 doesn’t just rush to give quick answers. Instead, it takes its time to consider each question carefully, trying to avoid mistakes by essentially fact-checking itself. This involves thinking through tasks step by step, which often takes some time, especially for complex questions.
DeepSeek-R1 has been shown to perform as well as OpenAI’s o1 on two important tests: AIME, where AI models judge other AI models, and MATH, which involves solving word problems.
However, DeepSeek-R1 has certain flaws. For example, it has trouble with simple games like tic-tac-toe, a problem also seen with o1. Additionally, there are concerns about its security; people have found ways to make it bypass its safety features, like when someone got it to describe how to make methamphetamine.
Moreover, DeepSeek-R1 avoids answering questions that might be sensitive or controversial in China, like topics related to political figures or historical events, showing that it’s designed with certain restrictions to comply with local regulations.
Parts of the DeepSeek code are on Github.
Origins in automated stock trading
A Financial Times story (unpaywalled copy) published earlier this year revealed that the developer of DeepSeek is High-Flyer Capital Management, a Chinese quantitative hedge fund that is a big player in China’s financial sector.
The fund started exploring the potential of AI for automated stock trading. “AI helps to extract valuable data from massive data sets which can be useful for predicting stock prices and making investment decisions,” said one of the fund’s managers as reported by Financial Times.
The fund is one of six Chinese groups with more than 10,000 Nvidia A100 processors, often believed to be a computational threshold for self-training large models.
“High-Flyer aims to achieve ‘superintelligent’ AI,” TechCrunch reports without mentioning the source of this information.
Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.
0 Comments
0 thoughts on “DeepSeek R1 reportedly thinks and reasons like OpenAI o1”