Answer AI
Uses an LLM-based AI model to generate an answer for the question asked.
Available on the STARTER plan and above.
The Answer AI action block is used to generate an answer to the user's question from the trained data.
When using the answer AI block, the user submits an open-ended question, which is then processed by the AI Studio to generate a relevant answer.
ZyndraAI/Talkie AB uses OpenAI GPT APIs to generate an answer.
Process in which the Answer AI works,
User asks a question
AI generates the answer from a knowledge base
Step 1: Define the question
This message will prompt the user to ask their question.
For example, your message could be,
👋 Hey! I'm an AI assistant, how can I help you?
Step 2: Choose a knowledge base
Choose the knowledge base from which AI will generate an answer.
Step 3: Configure your AI for accurate answers
LLM models require guidance to be able to generate a relevant and accurate answer. There are some tools you can use to guide your AI; they are:
Instructions/Prompt
A prompt is nothing but an instruction that helps the LLM know what to remember and follow while generating an answer.
The more clear, concise and brief your instruction, the more accurate your answers will be from the AI.
Things to write in your prompt:
Objective - What is its objective
Output format - Typically would be HTML/markdown.
Writing style - How should the answers be written.
Don'ts - Clear instructions of what to avoid.
Examples - Examples of question and answer.
LLM model
Select the LLM model that will be used to parse your training data and generate an answer.
Currently, ZyndraAI/Talkie AB only supports the following models:
gpt-4o
gpt-4-turbo
gpt-4
gpt-3.5-turbo
Restrict answer size
The maximum length of the answer is defined in tokens.
Tokens can be thought of as pieces of words, and 1 token=4 characters. Learn more
Creativity in responses
You can control whether you would like to receive unique responses every time a question is asked or get the same response every time the question is asked.
Define a number between 0 and 1. Nearer to 1 gives you distinct responses; nearer to 0 gives you fix responses. By default, we keep the number as 0.5.
Chunks
When the AI is asked a question, it first fetches relevant content from the knowledge base, which is then sent to the LLM model to generate an answer.
The more relevant content the AI can fetch, the more chances of accurate responses are.
The number indicates the number of relevant responses which will be fetched from the knowledge base.
Context length
Sends the recent sets of question answer pairs to provide the LLM more context of the current ongoing chat.
Benefit of this is that you can ask a follow-up question as 'its pricing', and the LLM would be aware that your previous question was about ZyndraAI/Talkie AB, so you meant "ZyndraAI/Talkie ABs pricing'.
Ask for feedback
For every AI generated response, we allow the user to submit feedback about the generated answer.
This helps you know how helpful are AI's answers.
Answer source
Displays the URL of the content source from where the answer was referenced from. This only works for knowledge bases built using a website URL.
Last updated