What are Large Language models?
AI tools like ChatGPT are Large Language Models that are a type of machine learning model which can read and respond using natural languages. These models are trained using self supervised learning in which enormous amounts of text are used for training. The AI models then comprehend human language semantics and understand user’s questions and respond to it.
What is DeepSeek?
DeepSeek is an Open Source LLM (Large Language model) like ChatGPT that was created by a
Chinese firm called High-Flyer. In November 2023, the first model of DeepSeek called as DeepSeek Coder was released. After many more versions, on January 20, 2025, the first free chatbot app called as DeepSeek- R1 model was released for iOS and Android platforms. Within 7 days, DeepSeek surpassed ChatGPT as the most downloaded free app on iOS in the US.
But how is DeepSeek different?
DeepSeek creators stated that it exceeded the performance of other LLMs such as OpenAI’s LLM and could perform at a much lower cost.
This caused a huge stir in the technology world. So, lets looks at how DeepSeek is different from other LLMs.
Mixture of Experts (MoE) Approach
DeepSeek uses a model called as Mixture of Experts that makes it different from other LLMs. Most LLMs use a traditional model that takes a user’s query and sends it to a massive network of processes that can involve looking through zillions of data and needs high power for processing. In a Mixture of Experts model, the users query is sent to different components called as experts and each component will perform a specific task. MoE method allows trainers to train specific experts on a particular domain or knowledge area. This splitting of tasks reduced the need for large computing power and lowers costs. However, it is important for correct routing of the user prompt to relevant experts to ensure quick and accurate answers.
Speed
The MoE approach makes the response quicker at a lower cost. This is because the prompt is split and sent to relevant experts that makes it quicker than other models. Splitting the prompt can make the inferencing process faster and make responses quicker. But, if only a few experts are reused repeatedly, that can place a large burden on the few while other experts are idle. This can reduce speed. This is why the DeepSeek LLM created a massive interest across the world.
How can we access DeepSeek?
We can download the DeepSeek App on iOS or Android, sign in with an email account and then you can start chatting with DeepSeek. However, there are restrictions that DeepSeek is placing now on new users signing up from outside of China. This is because of large-scale malicious attacks on the DeepSeek systems, there is a restriction on new users to sign up to DeepSeek. Existing users can however use it without any restrictions.
Restrictions on DeepSeek by countries across the world
Countries such as Italy and Taiwan have banned the use of the DeepSeek for its citizens while countries such as US, India, and Australia have currently banned its use for government employees due to the fact that users’ data are stored in People’s Republic of China.
Finally...
As modern technologies keep evolving with time, it is essential to keep pace with the developments to be more informed and efficient. While all AI tools aim to improve our tasks and knowledge, it is important to understand the advantages that each tool provides. While there are always some factors such as data privacy that we have been careful about in using the AI tools, informed careful usage will make our lives easier.
AI tools are a double-edged sword, as we can enjoy the benefits but need to be careful about appropriate usage.
Author: Priya H, Computer science teacher in HFS international
References:
Comments