@logankilpatrick Hi Logan,
The new turbo model is amazing. Here is a suggestion for future consideration.
Sometimes we chain prompts together. For example, we may ask GPT to do a task, and then follow that up with a prompt that further refined the output. We may do this a couple of times.
A really cool feature for future consideration would be the ability to feed the API a series of “user” role requests (without the intermediate “assistant” part)
Then you could let the AI generate the first assistant response, and then immediately reprocess the output by applying the next user prompt in the initial request. You could still charge for the output tokens. (and stream the outputs as they were processed - as we still want all the output)
It just saves two things:
- The developer doesn’t need to resend the “assistant” text and make another API call to carry out the next step in a sequence.
- Because of this, it will save tokens (for repeating the “assistant” text), and I imagine, it may reduce the number of API calls made on your servers. (latency etc)
To give you an example: (Not my actual use case)
- Generate an outline for a blog post with an introduction, two subtopic headings, and a conclusion
- Now generate a paragraph for the introduction
- Now generate a paragraph for the first subtopic
- Now generate a paragraph for the second subtopic
- Now generate the conclusion
There are much better use cases than this - but everyone will relate to this