#how to discourage bad chatbot responds and best practices

16 messages · Page 1 of 1 (latest)

earnest gate
#

Can somebody help me with making a chatbot?
What's the best POV for adding behavior
1.
is adding the behavior in 1st person better than 3rd person or vice versa
E.g. :
you are chattie an AI chatbot who chats like a teenager
Or
the following is a conversation with chattie, an AI chatbot that chats like a teenager
Or
chattie is an AI chatbot that chats like a teenager

I'm using Python, but the same issue will likely persist regardless of the programming language. I'm using an array to store the conversation and resend it in the following format:

Human: [user's oldest message]
AI: [AI's oldest message]
And so on...

However, when the bot's response is incorrect, it gets stored in the array and resent, which encourages the AI to respond in a similar manner. This results in a feedback loop, so the next responses will also become wrong.

tough stirrup
#

So my idea is to use fine tuned models

earnest gate
tough stirrup
#

Well you could try adding pre prompts if you're on budget but it couldn't fit your own needs if you require specific characters

#

Like the prompt is
"
I want you to act like a friendly guy:
Me: hey bro what's up
You:
"
Then the ai will generate after the "you:"

tough stirrup
#

Without fine tuning

#

So good luck with ur tight budget and requirements

woven hare
#

of course it will lead to a problem

#

it may not work properly when the prompt reach the max limit as it’s becoming longer

#

so you need to only add necessary information to the prompt

#

you can view here

#

I show a basic way to use text embedding to make semantic search