The Chinese lab DeepSeek has introduced a new AI “thinking module” that can compete with OpenAI o1. According to the developers, DeepSeek-R1 is capable of effectively self-verifying by spending more time reasoning through queries. This helps avoid common mistakes that occur in other models.
🚀 DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power!
— DeepSeek (@deepseek_ai) November 20, 2024
🔍 o1-preview-level performance on AIME & MATH benchmarks.
💡 Transparent thought process in real-time.
🛠️ Open-source models & API coming soon!
🌐 Try it now at https://t.co/v1TFy7LHNy#DeepSeek pic.twitter.com/saslkq4a1s
Like o1, DeepSeek-R1 performs tasks by planning and executing actions that help reach an answer. Depending on the complexity of the question, DeepSeek-R1 can “think” for several dozen seconds. DeepSeek claims that their model matches OpenAI o1 in performance on two popular tests — AIME and MATH.
Users who have already tried the new model note that DeepSeek-R1 can be easily tricked, ignoring safety mechanisms. One user received a recipe for prohibited substances from the model. In addition, DeepSeek-R1 blocks queries considered politically sensitive, such as questions about Chinese leader Xi Jinping or Tiananmen Square. This is likely a result of pressure from the Chinese government, which requires models to comply with “core socialist values.”
DeepSeek plans to open source DeepSeek-R1 and release an API. This is supported by High-Flyer Capital Management, a Chinese hedge fund that uses AI for trading decisions.
