Okay, so I’ve been messing around with this “karina and katrina” thing, and let me tell you, it’s been a ride. I basically wanted to see if I could get two different AI models working together, kind of like having a conversation between them. Sounds cool, right? Well, it was harder than I thought.
First Attempts…
First, I tried to just throw them at each other, like “Hey, Karina, talk to Katrina!” No dice. It was a mess. They just spat out random stuff, totally unrelated. It was like watching two toddlers babble at each other – cute, but not helpful.
Getting a Bit Smarter
Then I figured I needed to give them some structure. So I started by giving Karina a prompt, like “Tell me a story about a cat.” She did her thing, came up with a pretty decent story. Then, I took that story and fed it to Katrina, saying “Okay, Katrina, now make this story into a poem.”
- Karina: Story generation.
- Katrina: Poem adaptation.
This worked slightly better. Katrina actually tried to make a poem, but it was still pretty rough. It felt forced, you know? Like, it wasn’t a natural conversation.
The Real Pain Point
The biggest problem was getting them to understand the context. It’s like they had amnesia after every turn. I’d have to keep reminding them what we were talking about. I spent a lot of time just trying to keep them on track, re-feeding information, tweaking prompts. It was a lot of manual work.
Some Tweaks and (Slight) Success
I started playing with the parameters, you know, the temperature, the top-p, all that stuff. I’m no expert, but I found that lowering the temperature a bit made them a little more focused, less likely to go off on wild tangents.
I also experimented to give very specific instruction to one model. For example, when I use Katrina, I said:”You are good at rhyming, so please make any sentence into a beautiful short poem”.
Still a Work in Progress
Honestly, I haven’t cracked the code yet. I’ve gotten some semi-coherent results, where Karina writes a paragraph and Katrina turns it into a somewhat-related poem. But it’s still not what I’d call a “conversation.” It’s more like a very slow, very clunky collaboration.
So, yeah, that’s my “karina and katrina” experiment so far. It’s been frustrating, but also kind of fascinating. I’m learning a lot about how these models work (and don’t work). I’m definitely not giving up, though. I think there’s something cool here, I just need to figure out how to unlock it. Maybe I’ll try a different approach next time, like focusing on a specific topic or using a different pair of models. Stay tuned!
