O1 requires a developer or platform to host functions that call the tools. Each step requires a separate api call, meaning the model has to request an action, wait for the result, and then call the next one. And you have to call this recursively over and over again. This makes the barrier to using it higher, since you must provide the underlying function execution architecture.
Autonomous nodes do this automatically in the background.
How does o1 compare in cost and speed?
Let's say you're using a llm or ai agent to help with uganda phone numbers hubspot tasks, like chatting with website visitors as a way to do ai lead generation.
If you asked o1 to fill out a lead capture form in hubspot - that is, assuming you could connect it without a third-party connector - it would take 12 seconds and cost 5 times more than if it were done in botpress.
Part of the speed is due to the schema-based design of our platform, which means the system knows in advance what data is needed and what the structure of that data is like.
So, when you work with a tool like hubspot or salesforce, the system already has a predefined json schema that indicates what inputs are necessary, such as name, email address, company name, etc.
This allows autonomous nodes to reason through the entire task from the beginning, without having to stop and ask for more information at each step.
You can reduce the speed and steps by providing the context yourself, but at the cost of spending more time creating an o1-based application. So it is also about increasing the maintenance cost.
You mention greater reliability. What makes autonomous nodes more reliable than o1?
The most important way to increase the reliability of the results was to mix the chain of thought reasoning with the call to the protocol tool we created.
What we did was essentially create an environment where you can think - and all in code, not plain text. Honestly, it's faster and more reliable than the openai version, for several reasons.
One is that he can speak simultaneously, throughout the entire sequence of actions. That's something the openai model doesn't do.
Another reason is that the AI knows the output scheme. It knows what type of json it should spit out based on the query it receives.
And it is something that the developers themselves can adjust. If they use it and receive a lot of similar questions from users, which is the case for most, they can adjust the AI so that it always offers the best and most appropriate answer.