January 30, 2025:
Ai2 Outsmarts DeepSeek with Open Source Model - Ai2, a Seattle-based nonprofit AI research institute, has introduced Tulu 3 405B, an open-source AI model exceeding the performance of DeepSeek V3, a prominent system from Chinese AI firm DeepSeek. With 405 billion parameters, Tulu 3 405B also outperforms OpenAI's GPT-4o on certain AI benchmarks.
Utilizing reinforcement learning with verifiable rewards, Ai2 has boosted Tulu's capabilities, leading it to surpass competitors like Meta's Llama 3.1 on tests such as PopQA and GSM8K. This development underscores the U.S.'s ability to produce competitive open-source AI, available through Ai2's chatbot and platforms like GitHub.
