Frontier Harness, Local Models
Assuming a local model has sufficient permissions, it can do almost anything on your computer if it is guided wisely. The catch is that smaller local models still struggle to figure out complex tasks on their own, while running a frontier model for every single task gets expensive quickly (and you may not want to give a cloud model full access to your machine anyway).
The solution is to get a frontier model to develop a harness and skills (clear instructions, scripts, and guardrails) for the smaller model to run on. You use the smarter model once to build the rails, then let the local model do the actual work privately and at a fraction of the cost.
- #ai
- #llm
- #local-models