Prompts and GPT-3

I have been working with GPT-3 for a while now and have had access to the API for weeks. One thing I've found of real importance is writing prompts.

It's important to write prompts that provide enough context about the output expected and are also relevant to the environment. I tried lots of things, and it is definitely not easy to create prompts for the API. You really need to optimize a lot to efficiently receive outputs — and to make sure you don't end up wasting the tokens you have.

Providing the API with examples works well, but you really need to focus on writing good ones. I've seen that it's hard to train the API for specific and complex use cases. For example, the one I am working on: I am using GPT-3 for an app called Mailroo, and my motive as of now is to write good cold emails and sales emails. I definitely need to niche down and select a small, focused set of markets in order to successfully gain users and make something worthwhile.

I tried several sales templates, and it seems I still have a long way to go in writing good enough prompts for GPT-3. I'll keep improving them to get better email outputs from the API. I think I am at least near a point where the API will provide good enough email output for most instances. Let's see — hoping for the best.

---

Here are a few interesting examples I tried with GPT-3:

Also, there are going to be some interesting updates related to Mailroo — stay tuned.