🕶️

I tinker, therefore I am...

DeepSeek-R1 first look

DeepSeek-r1 is contemplating almost infinitely, like an introvert before making a phone call. It’s almost like it’s thinking aloud.

This helps to understand where the answer comes from and how to improve it. “Everyone” is currently talking about it, so I had to check it out too.

On reasonable home equipment, it is possible to test the DeepSeek-R1 (Qwen or Llama) with basic parameters: 1.5B, 7B, or 8B.

I posed a philosophical question: do staircases go up or down? Gemma 7B stubbornly claims they definitely go up.

Gemma 7b being confident

After a great internal debate, R1 concluded that — wait for it - it depends.

Deep Seek R1 thinking deeply

When I tried to convince the model to shorten the process to a minimum, it stated that actually, they only go down.

Initially, it looks great, but for more serious conclusions I need to work a little bit more with it.

— Jan 27, 2025