In this tutorial, we will provide you with some tips and suggestions to reduce the use of AI tokens to get better content at a lower cost.
Before we start #
To contextualize the information, we are going to provide, we will show a table with the usage of OpenAI tokens on our platform in the last week:
When you use a provider like OpenAI, charges are made for both entered text and generated text, unlike traditional writing assistants that resell you access to GPT-3 and only charge you for generated text. However, to keep their costs from getting out of hand, they set internal limits, such as using lower capacity models for short form templates or setting look-back character limits.
By not having these limits in our platform, it can be easy to overflow in the use of tokens, since GPT-3 (DaVinci) allows you to process (between prompt and completion) up to approximately, 16000 characters. We have detected that many users have used up to 15000 input characters to generate only 200 or 300 output characters. These values make the ratio per generated word increase the total average. Some users who use their tokens efficiently have costs per word generated in DaVinci up to 10 times less than the overall average.
However, the overall average per word generated using DaVinci on our platform is almost 10 times less than other offerings on the market. And if we average with what is generated with the Curie model, the costs can be as much as 100 times less.
Still, we believe you can save much more if you follow our advices.
Evaluate using a cheaper model when creating your short form templates #
While DaVinci is currently the most powerful OpenAI model, you can get just as good results using Curie if you include more examples when creating your templates. You will use more tokens at the time of generation, but your savings will be considerable, since their cost is 10 times less. In our ready-to-use templates section, 80% are developed with Curie and the results have been extremely satisfactory.
It is not always necessary to include all the previous text to generate more content #
Unless you are working on a creative story, where all the text defines what should follow, it is not necessary to include all the previous text as input. In most articles, 1 or 2 paragraphs of reference is enough for the AI to continue writing smoothly.
In our editor, you can add three asterisks at a certain position to exclude all content above them. This way, you will still be able to create high-quality content with fewer input tokens.
Generate several outputs in a single request #
When using templates or even in the document editor, we recommend generating multiple outputs in each generation, as you can save tremendously on the use of tokens, especially when the input text is long. For example:
If we generate content using a template that has 1000 input tokens and 200 output tokens and we use it to generate content 5 times, but one at a time, the expense in tokens is as follows:
- 1000 + 200
- 1000 + 200
- 1000 + 200
- 1000 + 200
- 1000 + 200
Total: 6000.
On the other hand, if we generate several outputs from a single request, the token usage would be:
- 1000 + 5×200
Total: 2000.
As you can see, generating multiple outputs can help you save a lot more.
Note: This option does not apply to the Muse provider which works differently in generating multiple outputs.
When using DaVinci, take advantage of the use of commands #
One of the best features of DaVinci is the ability to understand instructions, which allows you to get better results with fewer inputs. Thus, we recommend you to use commands with multiple instructions to generate them in a single request. For example: “Write a blog article about digital marketing that includes an introduction, the main related concepts and their importance. At the end include a conclusion that summarizes all the points mentioned”.
We hope the tips will help you to continue creating high quality content, but with an optimal use of your tokens to greatly reduce your costs.
