Being able to change the AI model for the Conversation AI.
V
Vivien Ducros
It would be amazing to be able to choose a smarter model in Conversation AI, I feel the model used is a bit dump (sounds like GPT 3,5). I'm ready to pay more for being able to choose a smarter model.
Log In
E
Eddy CEO WhatSnap.ai - Connect YOUR phone to GHL
Wait, we're already on GPT-4o? The responses still feel pretty robotic compared to what I get from Claude in other tools. Maybe it's the prompting or constraints, but yeah the personality issue is real.
S
Sales & Marketing
Merged in a post:
Upgrade Converstation AI Chat to GPT4.0
R
Richard Quinn
I understand that Conversation AI use GPT3.5 - the responses often miss context. Can we have an option to use GPT-4.o?
NOTE: this is available in Workflow Automations - see https://ideas.gohighlevel.com/automations/p/gpt-4-for-workflow-ai
a
arOS Soluções
Please! GPT is dull as a rock. If we're gonna pay per token, why not just connect all basic APIs (GPT, Claude, Gemini, etc.) and let us choose which one to use? GPT is WAY too limited. For chats it's super awful. Doesn't really follow proper instructions. Can't give a personality to it so it's not those dull answers
A
Abhishek Kumar
Vivien Ducros we are using GPT 4o , can you tell me what is the issue you are facing , the quality issues we can discuss
F
Filip K
Abhishek Kumar 4o can't improvise well and is not good at connecting answers well even if you tell it how to do it
A
Abhishek Kumar
Filip K we are upgrading it to 4.1 . We are testing it
a
arOS Soluções
Abhishek Kumar could I test different models for my Conversation AI agents? GPT 4.1 is pretty bad. Since I'm paying per token, what difference does it make to you guys?
(Unless you got openai credits, then what we pay actually goes as a full profit to you. Notwithstanding that, there should be no reason to NOT offer all LLM models for this)
It even shows here different models:
A
Abhishek Kumar
arOS Soluções why is GPT 4.1 bad? where is it struggling can you tell me . want to see the scenarios .
Anyways multiple AI models are coming very soon , We are introducing GPT 5 . mostly probably next month
a
arOS Soluções
Abhishek Kumar multiple models should solve the issue! I don't mind paying more (since I'm paying per token anyway) for a better quality output. Many times I direct it to behave or respond in a certain way, and it totally ignores it.
Is there a bigger system prompt you guys embed into it? Cause it's WAY harder to steer the AI inside GHL when comparing to other places. It doesn't really follow instructions to the letter… it kinda "guess" in between what I wrote
O
Omar Anaya
conversation ai is hallucinating a lot, and not sending the questions set on the workflow, instead is just sending some system messages to the leads.
A
Abhishek Kumar
we are using GPT 4 o
R
Richard Quinn
Abhishek Kumar - I have been trying to find any documentation on the GPT version used in Conversation AI, its not in the Help docs - and there are ideas still open asking for upgrades see https://ideas.gohighlevel.com/conversation-ai/p/gpt-4 - Can you add to documentation page to confirm this is in Conversation AI as well as Workflow AI. Thx
R
Richard Quinn
Abhishek Kumar - I'm still gettting conflicting opinions on this... Responses from the Community saying its GP3.5 Turbo ?
R
Richard Quinn
@Abhishek Kumar - and GHL Support Chat say GPT4 Turbo, so very confused now...
A
Abhishek Kumar
Richard Quinn Being the product manager for conversation AI I can confirm this , We currently use a combination of GPT 4O, GPT 4 turbo and GPT4 to power the Conversation AI bot. Utilizing a combination of these different models based on logical reasoning, intent detection, speed and hallucinations