Caveman Make Loop: Model Harness Different

TL;DR: Running Recursive's new Flint model in Oh My Pi (OMP) was a looping disaster of telegraphic "caveman speak." But running the exact same user prompt in OpenCode was night-and-day articulate and competent. The contrast sent me digging through session traces and harness prompts—and reconsidering how quickly I'd judged the model. Photo by Alex Moliski on Unsplash Two harnesses, one model At work, I just got a new MacBook Pro with 64GB RAM and an M5 Max CPU. Just in time for news about M6 CPUs, but oh well. Still, I'm told this should be pretty capable at running local AI models. So, lately, I’ve been jumping between two harnesses: OpenCode and Oh My Pi (OMP). Both seem capable, but have different designs and flavors. OpenCode feels like a structured, modular developer workbench. OMP aims at raw terminal speed, token frugality, and minimal ceremony. I haven't done an exhaustive code comparison, so these are just vibey impressions. At the same time, I’ve been running Flint, an…

Commenting on Bubbles has been removed. Why? If you want to talk about this post, write to the author, use the comments on their blog if it has them, or reply with a "RE:" post on your own blog. Suggestions? feedback@bubbles.town