The DeepSeek controversy has recently shaken up many, with events outside the AI scene often being interpreted as “USA vs. China”. In our view, this is wrong – the innovation that made DeepSeek R1 possible could just as easily have been created in northern Italy or Canada. It was a logical continuation of research into the use of reinforcement learning (RL) to train large language models. Amusingly, Nvidia (whose share price has suffered greatly from the DeekSeek event) has already achieved great success with RL approaches: Their Nemotron was already a benchmark topper in October 24, even though the model only has 70 billion parameters.
Rather, the key debate is closed-source vs. open-source AI models. This public discussion has often been misrepresented and overshadows the real issue: the importance of transparency and collaboration in AI development. DeepSeek R1 as an open-source AI model has proven to be groundbreaking by challenging traditional approaches and highlighting the transformative potential of open-source innovation .
The real debate: closed source vs. open source
AI models are often divided into closed source, such as OpenAI’s ChatGPT, and open source, such as Meta’s Llama 3, which is often used as the base model for newer open source models. Closed-source models keep their data and algorithms secret, while open-source models promote transparency and innovation. DeepSeek R1 stands out by embracing open-source principles without being based on Llama 3 and allowing global developers to access it freely. DeepSeek R1 has also been trained with a novel reinforcement learning (RL) approach that improves its adaptability and reasoning capabilities. This innovative approach is described in detail in this paper and sets it apart from the competition.
DeepSeek R1: A unique approach
DeepSeek R1 stands out in the AI landscape by not following the path of Meta’s Llama 3. Instead, it has created a novelty by incorporating reinforcement learning (RL), a method that improves its learning and adaptability. This approach is described in detail in this paper, showing how DeepSeek R1 achieves impressive reasoning and math capabilities. Its open-source nature means that developers worldwide can access and build upon its framework, fostering a collaborative environment .
The irony of Nvidia’s Nemotron
In October 2024, Nvidia released Nemotron, a robust mid-range open source model that was among the top models in many benchmark rankings. Ironically, the launch of DeepSeek R1 led to a sharp decline in Nvidia’s stock, despite its own earlier contribution to the open source movement. For more insight into Nemotron, check out our blog post. This situation highlights the unpredictable dynamics of the AI market and the growing influence of open source models.
Our approach: Integration of DeepSeek R1
At Evoya AI , we recognized the potential of DeepSeek R1 and integrated it into our model suite. This addition allows our users to explore and compare its capabilities with other leading models such as ChatGPT and Claude. We invite you to try DeepSeek R1 and experience its unique features first-hand. Sign up here.