It works great, especially with gpt4o-mini. I found out that the classification improves when I add a "Do Not Trigger for: " prompt to the intent classification function. Here is an example: d: Trigger FAQs when the user specifically asks for frequently asked questions or general information about car share. This covers: - Direct requests to see FAQs - Asking to return to the FAQ section - Seeking general information about car share steps Key phrases: FAQs, frequently asked questions, general information, common questions Do NOT trigger for: specific questions about insurance or bookings a:FAQs
Yes we've done something similar for other projects! Especially for confusing classes, in the blog post we reference Dental vs Doctor office intents www.voiceflow.com/blog/benchmarking-hybrid-llm-classification-systems
Amazing feature. Up until now I built a sophisticated Component to do intent classification through GPT3.5 / 4 This changes everything and simplifies my flows. Thank you for this killer feature
Nice work! Might be cool to add the intents creator (you used Claude) inside Voiceflow so people don't have to go to another area :). Love this tutorials!
The reason why "Correctness is a life or death situation" works is because it is trained on us humans. I've been using it in my prompts and seen a good ~15% increase in output performance
It works great, especially with gpt4o-mini. I found out that the classification improves when I add a "Do Not Trigger for: " prompt to the intent classification function. Here is an example:
d: Trigger FAQs when the user specifically asks for frequently asked questions or general information about car share. This covers:
- Direct requests to see FAQs
- Asking to return to the FAQ section
- Seeking general information about car share steps
Key phrases: FAQs, frequently asked questions, general information, common questions
Do NOT trigger for: specific questions about insurance or bookings
a:FAQs
These are great prompt tips! Thanks for sharing!
Yes we've done something similar for other projects! Especially for confusing classes, in the blog post we reference Dental vs Doctor office intents
www.voiceflow.com/blog/benchmarking-hybrid-llm-classification-systems
Amazing feature. Up until now I built a sophisticated Component to do intent classification through GPT3.5 / 4
This changes everything and simplifies my flows. Thank you for this killer feature
Nice work! Might be cool to add the intents creator (you used Claude) inside Voiceflow so people don't have to go to another area :). Love this tutorials!
I agree this would be a cool feature :)
The reason why "Correctness is a life or death situation" works is because it is trained on us humans. I've been using it in my prompts and seen a good ~15% increase in output performance
This is a great feature! I will admit though, the first thing I did was go to my NLU intents and look for the "download utterances" button 🤣
My lord this is so goood!!!
Thank god for bringing this.
This was so needed!
Pretty cool feature! 🎉🎉🎉
Great video! Where can I find the link to the Antrophic template?
Really cool feature :)
So how do we change the prompt wrapper, where does the action and descrption go?
Great feature. Can we get the prompt template please? I don't see it in the description or comments.
We just added it to the description -- sorry about that! developer.voiceflow.com/docs/intent-classification-using-llms
Interesting but I don't see the setting button you used, could it be that this feature hasn't rolled out or is it just for the Pro tier? Thanks
Try refreshing! It launched 10 minutes ago.
@@Voiceflow That worked. Thank you
Where is the prompt template mentioned by Pete at the end?
Hi! We added it into the description, but you can find it here! developer.voiceflow.com/docs/intent-classification-using-llms
This is good are intents capped at a certain amount
Mas que merda. Mudou o que? Nada. Nao melhorou em nada e o campk de descricao pré-existente
Make sure you turn on the LLM classification!