How do content-creating platforms such as copy.ai and jarvis.ai, make their models?
Do they use the playground and then call the API for the playground ?
Even if they use fine-tuned models, do they use the playground in any domain?
That’s kind of how I do it in my app - CopyAim.com
I don’t use fine tuning yet because the regular way works amazing enough.
Additionally I don’t think I will ever be using fine tuning (at least in a current state of things).
Playground + HIGHLY customizable prompts + API is a way to go.
In Copyaim.com I also have a few experimental features that use 3-5 different API calls, building things contextually on top of each response.
2 Likes
Fine tuning is a great way to reduce size of prompts. If you already make advanced prompts, you probably can save money by moving parts of that into fine tuning.
1 Like
That is true, but in my case the cost of prompts is literally next to nothing. I use 15x tokens just messing with playground than all my customers together (given we currently have around 1000 users and completely free unlimited generations.
Limiting the prompt size is good and all, but with a ~1000-1200 tokens curie prompt that contains 5-6 examples it can give me 10 end results (“n: 10”) and it still cost below $0.01
1 Like
Yes if your starting out people don’t make. much any way you are trying to get started up so all that should been put a way so you can keep building your company up.