How to set context for AI hints

When you start a brand new project, the machine learning model behind the AI text generation is pretty basic, and mostly based on the general knowledge that we extracted from your website and other data sources. You can choose which AI engine to use to transform this knowledge into conversations in the general settings of your project, choosing between GPT-3 by OpenAI and our built-in models.

How to set the context for the built-in model

You've already done this when you uploaded your knowledge assets. Once the training process is complete, we'll use that knowledge to suggest responses for every new message that your customer sends.

How to set the context for GPT-3

If you prefer to go with the big players, open the "Settings - Integrations" menu and click the "OpenAI" settings tab. The settings are pretty self-explanatory and you can set the context in plain natural language (see the example below).

You should provide a general description of the context of the conversations, and then provide 1 to 3 example questions (with suggested responses) that your customers may ask.

The "temperature" setting affects the flexibility of the model (the higher the temperature, the more improvisational the model becomes), and you can also adjust the maximum suggestion length.

How to fine-tune the model

Once the initial context is set, the learning process starts. Your virtual assistant will be using the conversations done by your human agents to fine-tune itself, and you will notice the change quite soon. On average, it takes 100 to 1000 live chat sessions to get enough data for the model so that it will be able to provide meaningful responses.

We can also help you with the fine-tuning process by custom-analyzing your data and suggesting the fine-tuning mechanics. Just ping us at with your account details, and we'll build a custom fine-tuned AI model for your customer service and live chat automation needs.

Last updated