GETTING STARTED WITH GPT-3

You’ve probably spent more time than you should toying with ChatGPT, but the GPT-3 engine behind it is a paid-for tool that offers a lot more flexibility, especially when it comes to writing apps and training the AI with your own data.

We don’t have the space to do a deep dive into GPT-3 here, but if you want to start exploring what this enormously powerful tool has to offer, you should start experimenting in OpenAI’s Playground.

The first thing to do is visit openai.com/api and sign up for an account. You can sign in with your Google or Microsoft credentials or register afresh with an email account. You’ll get $18 of free credit to play with, but you’ll need to enter payment details to start doing any serious experimentation. We’ll come back to controlling costs in a bit.

OpenAI provides a lot of great documentation and examples that it’s worth working through once you’re registered. However, the Playground

( beta.openai.com/playground) is the place to start cutting your teeth.

There’s an enormous amount to play with here, but perhaps start by using the preset dropdown at the top of the screen and selecting one of the options. For example, “natural language to Python” is a preset that allows you to type what you want a piece of Python code to do and let the AI do the coding. Or, under “more examples”, try “Ad from product description” which could help you sell an item on eBay by providing basic product details. There are at least 30 different examples to choose from.

On the right of the Playground screen are various parameters to play with. Again, we don’t have space to dive into them all here, but you can choose which language model you use (Davinci-003 is the most advanced, but also the most costly), while turning up the temperature allows the AI to be more creative in the answers it gives, but runs a greater risk of random or nonsense output. Again, read the documentation to find out what they all do.

Pay particular attention to tokens. Tokens are how OpenAI charges for the service. Tokens basically equate to the length of the output provided, so if you’re asking the AI to produce reams and reams of text, it’s going to cost a lot more than one-word answers. You can adjust the maximum length to keep costs down.

To ensure you don’t accidentally run up a massive bill (which is more of a risk when you’re running apps against the API), you can put a soft and hard cap on your monthly bills in settings. Click on your profile in the top-right of the Playground screen and select Manage Account | Billing | Usage Limits. Set a low limit (say $20) until you’re comfortable with how it’s charged. It will probably work out a lot cheaper than you fear, but it’s better to play safe to start with.

If your app suddenly stops working, go back into billing and check you’ve not hit the cap.