from zero to one

GPT-3 API alternative

Copywriting based on keywords

I would like to create sentences out of a few keywords, like "Intel, apple pie, spirit".

Something like textgenrnn does, or at least should do.

I am trying to create a proof of concept API. Let's see how far i get. ...

GPT-3

As GPT-3, which seems to be the no plus ultra at this time, is not publicly available. A good read about the official API is from Twilio here. You may also try to apply for an API key here.

All those new shiny GPT-3 based copywriting tools do not provide an API as well. I tried copy.ai as a benchmark, which has no such feature. Next I tried jarvis.ai, which cannot do this either.

GPT-J

And what I found so far is GPT-J. It looks at least promising, so let's try this out, and see how far we may get. I found as very frequently in an article on towardsdatascience.com. Although what does not sound very promising is.

Because GPT-J was trained on a dataset that contains GitHub (7%) and StackExchange (5%) data, it’s better than GPT-3-175B at writing code, whereas in other tasks it’s significantly worse.

You can play around with it, in this notebook.

There is also a self-hosted API available here.

A python package is also available here

I tried to test the model at eleuther.ai, which is incredibly slow and producers long text, which looks from my first impression more or less random.

GPT-Neo

Found this article

Best text generation API so far

So npcloud.io seems to have what I was looking for.

Pay as you go, starting at $1 a month.

Supports many models (GPT-Neo, GPT-J, etc. ...)