New AI and how to use Smitty

What we're working on

In the past period, we've had discussions with various companies interested in our tool. We are moving forward with new developments, such as making subscription plans available and adding a new AI as an option.

We've gained access to the Claude AI API. A significant advantage of this API is that larger documents can be sent in their entirety, improving the coherence of the chatbot, especially with more complex documents. We are working on integrating this feature into Smitty, allowing users to choose from an even wider range of AI models when creating a chatbot. The choices will include GPT 3.5, GPT 4, Claude, and PaLM 2.

Smitty Usage

Our advice is to proactively populate the knowledge base with well-structured documents within the character limits. These character limits can be set during the creation of the chatbot.

If you use automated document splitting, there is a risk of losing the meaning of a document segment, potentially leading to incorrect answers. This occurs because, depending on the chatbot settings, the AI may send either one snippet of 1000 characters or three snippets of 1000 characters each for generating an answer. Although these three snippets may be related, they might not necessarily have a strong connection. The AI will then combine the information to generate an answer, sometimes accurate and sometimes not.

If the AI lacks available information, it may still generate answers with nonsensical content. To address this, we have introduced a truth and facts temperature setting in the chatbot. This setting helps prevent inaccurate answers. Incorrect responses can result from insufficient documentation or improperly split documents.