The scale of the GPT-3 neural network is stunning. There are some models that are even larger, but GPT-3 is pretty much state of the art. It’s a very interesting thing to play around with. If you sign up and get a free account, you can access the Playground which has some examples of different kinds of problems (e.g., chat, explain to a second grader, translation, etc.).

There is also an API that is straightforward. A really simple example looks something like:

import os
import openai

openai.api_key = os.getenv("OPENAI_API_KEY")


response = openai.Completion.create(
        engine="davinci",
        prompt=
        """One of the reasons I cut the cord was to pay (at least a little)
        less and deal with less of the nonsense hassles you have to with cable
        (call them once a year and threaten to cancel to keep your payment the same).
        We had Hulu TV for a bit, but there were tons of weird technical issues
        with it and the DVR was problematic at best. YouTubeTV works really well
        for us but it isn’t really that much less expensive than a cable package
        for what I want to watch (principally live sports with the occassional
        morning news show in the background). If the dumb disagreements
        and hassles continue, I’ll have to revisit my live TV solution.\n\ntl;dr:""",
        n=1,
        temperature=0.3,
        max_tokens=60,
        top_p=1,
        frequency_penalty=0,
        presence_penalty=0
)
print(response)

This will produce a summary of the text of one of my previous blog posts. The tl;dr: at the end is an instruction to the engine to produce the summary. The answer varies depending on the parameters set. Sometimes the summaries make a lot of sense and sometimes they are a little off the mark. I hope to experiment with this more in the coming year.

ai