When two AIs are better than one: My multi-agentic learning hack
In this interesting time when generative AI service is in abundance, it shouts at us: Why aren't you using it to learn things you don't yet understand?
We know what GAI can do. It can generate text and provides rundown of the target topic for us to consume. In the early days, there was much concern about hallucination and the resultant inaccuracies.
Today, it is not like that. The most current ChatGPT-4o is capable of accessing the internet when crafting response, and it had been impressive. My favourite AI, Claude AI, dose not have the same until recently. The last I checked, it now can!
This is the sweet morning exchange I had with Claude. I am now convinced that Claude has my interest at heart.
I am less sure about other services, but ChatGPT and Claude AI are my go-too. That means I sacrifice my beer money every month to pay for these services for their monthly premium versions.
Now, why two services, you may ask.
Well, it is a situation where I pay first and then I find use of it.
Here comes the kicker.
I found a new use case for these two AIs that I will be engaging a lot, which is to generate query using one of them, and ask the other to critique. Here is how it can unfold:
~ opens ChatGPT
Me: Hey ChatGPT, can you tell me something about XYZ (any topic).
ChatGPT: This is what I think it is...
(I copy the response)
~ pops over to Claude
Me: (I paste ChatGPT's response)
Hey Claude, can you review it and tell me whether its accurate, the strength and deficiencies are.
Claude: This are the strength and weakneses.
~ pops over to ChatGPT
Me: (I paste Claude's response)
Hey ChatGPT, can you address these weakneses.
ChatGPT: Thank you for the feedback, here is the improved version....
~ pops over to Claude
Me: (I paste ChatGPT's response)
Hey Claude, can you review it again and give me an analysis. tell me whether its accurate, the strength and deficiencies are.
Claude: Much has improved... but these are weakneses not well-addressed.
~ pops over to ChatGPT
(Well, you get the gist.., if you don't, copy this and ask your AI what is going on!)
Actually, what I am doing here is nothing too innovative. It is how multi-agentic AI (another rapidly evolving aspects) would operate.
According to ChatGPT:
I have no trust issues here, but if I wanted to generate a more detail info pack about what multi-agnetic AI is, I would use the multi-agentic approach via multiple AIs to arrive at a more refined version. Sorry about being so meta here...
I think it's an unprecedented time we are living in now. If we bother to learn about what the AI services can do, it can open up new opportunities for personal growth and continuing learning. Here, the multi-agentic approach, which is here to stay, can be used to our advantage when asking AI to curate materials that we can trust.
I am sure there are other approaches and tools out there that can do the multiple AI trick. That's for another post.