By the way, you can extract JUST the reasoning from deepseek-reasoner, which means you can send that thinking process to any model you want before they answer you. Like here where I turn gpt-3.5 turbo into an absolute genius!

What's even crazier is that this is done with a max tokens of 1, meaning I am not even paying for any output here.

Everyone is getting ahead with AI. You should be too. Summarize documents, craft emails, and generate custom content instantly with this powerful tool. It's like having ChatGPT tailored for your job. Try it for free.

This is really interesting because: Sometimes a faster, cheaper model may not know how to answer a question correctly or imagine when you need to fix code. As long as the thinking process is correct, even a dumber, cheaper model may get the answer correctly.

So in theory.... 1️⃣ Use DeepSeek for reasoning to break down a prompt into steps. 2️⃣ Store and track progress using long-context models (Gemini/MiniMax-01). 3️⃣ Research individual steps using tools like Perplexity. 4️⃣ Combine everything for a genius-level final response! 🔥 Sounds like agentic reasoning?

🤯 So you're basically channeling deepseek’s brain into GPT-3.

Or any model you want, since I am using @OpenRouterAI

Deepseek + Claude Sonnet might produce some great results then?

this is like a form of speculative decoding isnt it?

It's fascinating how efficiency can come at a low cost. Who knows what other breakthroughs lie ahead? #Innovation

I need to try this. Can you please upload it to your GitHub?

After seeing several posts about watching R1’s reasoning process, I was just thinking about the use cases for watching it think through problems. Just had to keep scrolling to find your useful post.