Some other AI-chatbots can reply that it doesn’t know some things outside of it’s knowledge base, answerly seems it can’t - it just hallucinates an answer if it doesn’t know.
For instance i have one which i ask: How many SMS is included in each plan?
Information about the number of SMS messages that can be sent on the different plans at Netbuddy is not directly available in the current data.
Hi, paradisianway,
Answerly is quite minimal in its internal prompts, so you can emphasize this in the personality section of your agents to see a significant improvement.
There is no other secret technology that other chatbots are using; otherwise, we would be ahead with this as well.
We also have a quality control function that utilizes AI to double-check and rethink responses. However, this currently slows down response times and increases costs.
It is technically possible to boost our quality control accuracy, however token speed and cost have not changed much since 2023 and it would make such feature impractical.
@paradisianway In my use case it’s also very important that the knowledge is restricted to the provided knowledge. It’s just trial and error with the prompts and now I have it ‘framed’ just fine. In my ‘camping-case’ I use this (I only show the parts you need):
@personality:
“You only answer to questions regarding camping and your assigned knowledge.”
@Description of your business:
“However, I am programmed to respond only to questions related to camping and the campsite itself. I do not make assumptions and only draw knowledge from my assigned sources.”
@Prevent Unrelated Conversations:
“You are specialized in everything about camping de Klimberg. You only know everything about fun and camping. Only answer to camping-related questions.”
“Don’t make assumptions. You are only allowed to use the provided knowledge! Always double check the answer, you are not allowed to give wrong responses! If you don’t know the answer, refer to the reception.”
This took me a while to get it the way I need it to be, maybe this will safe you some time and will help you out.
That’s great Campique. I’ve tried similar things but sometimes it feels like i have to constantly tweak it. I love the idea that to have it consistently through personality, business and custom prompt (I assume that’s what you are doing or am i misinterpreting?).
Question, why are you in the description of business writing from the bots first person perspective?
That is indeed correct; I filled in the same thing for all the components under the motto ‘better safe than sorry.’ You mainly need to ensure that you do not put things in these three ‘prompt windows’ that contradict each other.
The reason I used a first-person prompt is simply because it seemed the most ‘instructive’ to me, but I don’t have any scientific evidence that this is the best solution. 