Combine DeepSeek with GPT-3.5 Turbo for Efficient AI Output

This is a pretty interesting technique I have seen a few times online today. TLDR; use deepseek to produce reasoning, use 3.5-turbo or similarly small model to produce output based on reasoning https://egghead.io/combine-deep-seek-r1-reasoning-with-gpt-3-5-turbo-for-the-cheapest-fastest-and-best-ai~24oy1

· Sorted by Oldest

This is a pretty interesting technique I have seen a few times online today. TLDR; use deepseek to produce reasoning, use 3.5-turbo or similarly small model to produce output based on reasoning https://egghead.io/combine-deep-seek-r1-reasoning-with-gpt-3-5-turbo-for-the-cheapest-fastest-and-best-ai~24oy1

· Sorted by Oldest

Anthony P.
·
Here's the code https://github.com/johnlindquist/deepseek-reasoning/blob/main/index.ts
Anthony P.
·
There is an interesting extension to this experiment that may be fun to explore: What if you let deepseek continue on with its output after generating reasoning, and simultaneously generate output with a cheaper model using that reasoning, and then Eval them by comparing final outputs
Anthony P.
·
Can see how close a cheap model can get to deepseek's output with the right reasoning fed into it
Parker S.
·
thats a cool application of the different models for sure

Anthony P.
·
Here's the code https://github.com/johnlindquist/deepseek-reasoning/blob/main/index.ts
Anthony P.
·
There is an interesting extension to this experiment that may be fun to explore: What if you let deepseek continue on with its output after generating reasoning, and simultaneously generate output with a cheaper model using that reasoning, and then Eval them by comparing final outputs
Anthony P.
·
Can see how close a cheap model can get to deepseek's output with the right reasoning fed into it
Parker S.
·
thats a cool application of the different models for sure