1. Reasoning model from china that was built with reinforcement learning and not supervised fine tuning which chatgpt was trained on

2. Reinforcement learning is a method used to train cars to drive themselves, robots learning to walk or do certain actions, its like you learning to ride a bicycle

3. It spends more time thinking about the answer and getting better and in depth answers every time.

4. Distilled models allow anyone to try it out on their own. 

 

Why is it better than ChatGPT?

1. On par performance in terms of math, science and coding with o1 reasoning model which is 200$ a month, Deepseek r1 is free

2. It's open source, you can run it offline locally on your computer

3. It shows you its thinking process, which o1 skips completely

4. Don't need to write the most in depth prompt. It's able to find the best outcome through reasoning

5. API is 27 times cheaper than o1, making it a no brainer for apps to use it over openai's models.

 

Also talked about:

1. DeepSeek Chat

2. Running DeepSeek-R1 locally on your computer with Ollama

3. DeepSeek on a UI: anythingLLM or LMStudio

4. DeepSeek r1 with Perplexity:  GOD Mode

 

You can build anything with DeepSeek-R1. Here's how:

1. Building AI Agents with DeepSeek-R1 & Browser Use

2. Deepseek Coding Assistant in VS Code & Cursor AI

3. Deepseek AI Automations: Make.com + Openrouter

 

Problems with DeepSeek-R1:

1. Your queries are stored in PRC

2. There's censorship for specific requests you have

3. Not the best when it comes to creative writing.