Building Your Own AI System | Getting Started With OpenAI Swarm In Python

[@jiejenn]

Any other topic you want me to cover? Let me know in the comments!

[@HerroEverynyan]

I've been having issues grasping the whole agent thing, but the way you explain it fills in the gaps. Thanks so much! It's so cool you explain things so very in detail.

[@jiejenn]

Awesome. Glad to hear the video helped!

[@HerroEverynyan]

@@jiejenn I also think that people would greatly benefit by such an explanation on DSPy. I've been struggling with it for a while now. I want to use it in production, but first I need to understand it well to create a prototype to pitch it to my employer. Otherwise we're just using plain old prompt engineering. And maybe I'm just dumb, but most tutorials don't clarify it for me.

[@mw-st3qm]

i think i still prefer magentic prompt_chain for now, but its cool to see another way to do this.

[@jiejenn]

First time hearing magnetic prompt chain, will have to check it out.

[@the_runofff]

do you think theres a big advantage to having a set of agents each with a logical set of functions to call? why not have a single agent with all the functions? then you dont need switching behavior

[@jiejenn]

When you have an agent with too many tools available, sometime it makes a bit difficult to manage + unwanted behaviors. The second reason using multiple agents is multiple agents can be ran in parallel to run different tasks simultaneously. And by allocating different tasks to each agents, you get a more accurate response and execution. I hope that make sense.

[@mw-st3qm]

its easy for 1 mega agent to have its context window overwhelmed. i encountered this when trying to add a function to my agent that needed to distill a summary from 32kB worth of xml data. so i think its one way to help mitigate this, as well as reducing the chances that your bot calls a function it shouldnt due to semantic similarities. i think its similar to how, when designing an API, you should only expose methods you want consumers to be able to call.