Chaindesk is powered by large language models (LLM) like OpenAIās ChatGPT.
As powerful as LLMs are, they can sometimes produce inaccurate or out-of-scope answers. This is called āhallucinationā.
Chaindesk provides multiple mechanisms/features to decrease hallucination.
Avoid duplicated content
Adding multiple times the same documents/information, even translated should be avoided.
Duplicated content can impact the Chatbot's performance.
It is best to add a piece of information only once, whatever the language
Ā
Use GPT-4
Using GPT-4 can help to generate more accurate and truthful responses.
This model is also more respectful of the instructions given in a prompt.
The downside of GPT-4 is that it costs more credits and is slower than GPT-3.5-turbo.
Ā
Fine-tune your Agentās answers
Use our latest feature, , to correct an Agent's answer.
Ā