Introducing the ChatGPT plugin in ONLYOFFICE Docs

Originally published at: https://www.onlyoffice.com/blog/2023/02/chatgpt-plugin-in-onlyoffice-docs/

Now you can get accurate answers to your questions, quickly find information, generate texts, and even get your code built with the help of the renowned AI helper, right in ONLYOFFICE Docs. Here’s everything you want to know about the new ChatGPT plugin.

Introducing the ChatGPT plugin in ONLYOFFICE Docs

About ChatGPT

An AI-based chatbot called ChatGPT was first launched by OpenAI, an artificial intelligence research laboratory, in November 2022.

We asked ChatGPT to introduce itself, and here’s what we got:

ChatGPT is an AI language model developed by OpenAI. It uses deep learning techniques to generate human-like text based on the input provided. It can be used for various natural language processing tasks, such as text completion, conversation, and language translation.

The model was trained on massive amounts of text information from the internet including books, websites, forums, Wikipedia articles, many other sources openly available in the web. To reach both the required level of understanding the human language and imitating it, the developers utilized supervised learning and reinforcement learning techniques with an assistance of human trainers.

What ChatGPT can do

ChatGPT can perform multiple tasks within responses that involve processing natural language:

  • Text generation based on inquiries;
  • Predicting and finishing pieces of text based on the input;
  • Question answering;
  • Translation;
  • Text summarization and structuring;
  • Sentiment analysis;
  • Conversation-style communication.

All the above mentioned skills make the chat convenient to use without significant effort. As responses to your inquiries, it generates text it deems the most appropriate to your request.

However, it is not limited to just chatting and answering questions. Possible scenarios go as far as creating academic texts, composing music, building code, translating content, writing fictional stories, and much more.

Here are a few examples of ChatGPT’s responses generated in ONLYOFFICE Docs:

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Using ChatGPT in ONLYOFFICE Docs

How to install

To install the ChatGPT plugin in ONLYOFFICE Docs cloud or on-premises edition, go to the Plugin Manager in the Plugins tab, find ChatGPT and simply click Install.

If you are using ONLYOFFICE Desktop Editors, first you need to get the plugin on GitHub and follow the instructions for ONLYOFFICE Desktop Editors to install it in your solution.

How to use

Open the plugin, find it in the Plugins tab and launch. ChatGPT interface will pop up in the left panel:

Introducing the ChatGPT plugin in ONLYOFFICE Docs

To activate it, you need to submit an API Key which you can obtain from OpenAI. Keep in mind that ChatGPT API is a paid feature.

Once the plugin is set, you can start using it. To send an inquiry, write your message in a free form in the text field and click Submit. ChatGPT will process the request and insert the response in your document in plain text:

Introducing the ChatGPT plugin in ONLYOFFICE Docs

Additional settings

There are additional settings available below the text field that can help you configure the output:

Model parameter allows you to select an AI model to use for processing your request. Available models have different performance speed and use different tokens to process requests on the backend.

Maximum length is a maximum number of response tokens to use in the answer. Tokens are not words or symbols, but rather are batches of characters that the system uses to estimate the volume of text. It is hard to define how many symbols a token combines, but usually it is around 3-4 symbols. For example, a word “hamburger” will be divided into three tokens: “ham”, “bur”, and “ger”.

Maximum length also includes the number of tokens used in the request. If your Maximum length is 2000 tokens, and your request is 200 tokens, the response will contain maximum 1800 tokens. You can see the real-time token count for your request below the text field.

Temperature defines how much randomness is present in the output, i.e. the amount of uncommon, “creative” decisions made by ChatGPT. The lower the temperature, the more common words and language forms will be used, and vice versa, where 0 is the lowest and 1 is the highest.

Top P is an alternative way of defining the randomness. It affects the word sampling in the answer, where 1 will give you the most commonly used words, and 0 will give the least common results. We recommend using only Temperature or Top P variable in each inquiry for the best results.

Stop sequences parameter defines the text information that will work like a stop word for Chat GPT. For example, if you set “carrot cake” as a stop sequence, it will stop generating text when “carrot cake” is the next word combination in the queue. ChatGPT will also not include it in the final text.

Get the ChatGPT plugin in the updated Plugin Manager

You can use ChatGPT in all editions of ONLYOFFICE Docs and in ONLYOFFICE Desktop Editors. The plugin is already available in the Plugin Manager of the cloud solutions located in the Plugins tab of the editors.

Introducing the ChatGPT plugin in ONLYOFFICE Docs

We also improved the Plugin Manager interface and mechanics, making it even more convenient to use. Here’s what’s new:

  • App search;
  • Categories for easy navigation;
  • Bug fixes and improvements.

Useful links

OpenAI website

ChatGPT plugin on GitHub

Stay connected with Zoom plugin in ONLYOFFICE Docs

How to publish your own plugin in Plugin Marketplace

Best ONLYOFFICE plugins for online educators

Really cool plugin, are you using a custom prompt to generate markdown or something similar so you can get the tables from it?

The used model here is actually GPT-3.5 / text-davinci-003 not ChatGPT, which is a different model. When ChatGPT will be released as an API, the integration can be used with the ChatGPT model, however currently this is not using the ChatGPT model, but the above mentioned one.

Also it should be mentioned that the validity of the information the model generates fluctuates between true and really really false. This should be mentioned when generating text with this model.