Secrets of DeepSeek AI model revealed in landmark paper
7 hours ago
- #DeepSeek
- #Machine Learning
- #Artificial Intelligence
- DeepSeek's AI model R1 excels at reasoning tasks like mathematics and coding, and is a cheaper rival to US-developed tools.
- R1 is an 'open weight' model, available for download, and has been downloaded 10.9 million times on Hugging Face.
- Training R1 cost $294,000, on top of $6 million for the base LLM, significantly less than rival models.
- R1 is the first major LLM to undergo peer-review, setting a precedent for transparency in AI development.
- DeepSeek used pure reinforcement learning to train R1, allowing it to develop its own reasoning strategies.
- R1 has been influential in AI research, inspiring reinforcement learning work in LLMs in 2025.