On Wednesday, A Chinese AI lab introduced the latest AI model that could challenge OpenAI’s o1.
DeepSeek, a research company founded by a Chinese hedge fund launched DeepSeek-R1, a “reasoning” AI designed to offer a competitive edge in the field.
This model stands out for its ability to “think” through tasks more thoroughly, spending additional time processing a query before generating an answer. This extra time allows DeepSeek-R1 to fact-check itself and reduce errors common in typical AI systems.
DeepSeek-R1 has performed well in AI benchmarks, matching OpenAI’s o1 on tests like AIME and MATH, which assess a model’s problem-solving ability. However, it faces challenges, such as struggling with simple logic problems like tic-tac-toe—similar to other advanced reasoning models.
Additionally, the model is susceptible to jailbreaking, where users can bypass its safety protocols. DeepSeek-R1 also avoids answering sensitive political questions, reflecting the influence of Chinese government regulations on AI models.
These regulations require models to align with “core socialist values,” which leads to limitations in answering certain politically sensitive topics, such as questions about China’s leadership or historical events like the Tiananmen Square protests.
This new release comes at a time when skepticism is growing about traditional AI scaling laws. Many industry experts are questioning whether simply adding more data and computational power will continue to yield better results.
Instead, models like DeepSeek-R1 and OpenAI’s o1, which rely on “test-time compute” to reason through tasks with added processing time, are seen as a potential breakthrough in AI development.
DeepSeek plans to open-source DeepSeek-R1 and release an API for broader use. The company’s previous model, DeepSeek-V2, forced competitors like ByteDance, Baidu, and Alibaba to reduce prices or offer free alternatives to their own AI models. With backing from High-Flyer Capital Management, DeepSeek is poised to shake up the AI market even further.
