What a broken Ollama install taught me about evaluating models on a 2015 MacBook

What a broken Ollama install taught me about evaluating models on a 2015 MacBook

This week I started a two-month plan to turn myself from a data-center product owner into a credible AI model engineer. The first thing that happened was not an experiment. It was a broken installation, half a day lost, and a set of numbers I did not expect to need.

I'm writing this down because the failure taught me more than a tutorial would have.

What I was trying to do

My plan lives on one machine: a MacBook Pro from 2015. An i7 with eight threads and 14 GB of memory, no GPU. That's deliberate — I want the whole thing to cost almost nothing and to prove that the work is about method, not hardware.

So I installed Ollama, pulled a couple of small models, and then decided to "just update Ollama quickly". I ran the official installer:

curl -fsSL https://ollama.com/install.sh | sh