I don't know how productive I'm being but I'm using Llama3 via Ollama on a M1 Mac. It's as good as Copilot and Gemini for most things and I'll use those models if I need a little bit more. I prefer the privacy of the local models. I use it both through the command line and with the Open WebUI web interface. I use it for programming tips, learning, research, and writing. As a simple example, I wrote a (reusable) prompt for doing Chicago style title capitalization a few minutes ago. Normally I'd have to search for a web based tool and then manage through the crap. It's much quicker to ask a local LLM.

Similar to another reply, my M2 16GB MacBook happily runs Mistral 7B, and it will do a decent job of most requests.

I use it where I’m dealing with internal code or data for work where I need to know it’s all staying on-device.

I’ve had it roughly translate portions of code to another language (normally one I’m very familiar with so I can vet it), create mermaid syntax flow charts for code where I need to visualise a process for non-technical consumption, and compare two very similar job descriptions to understand where a candidate might be better for one role or another.

I have on occasion also asked it to condense a wordy email going out to senior staff but I find Mistral 7B is a bit all or nothing and will take my 5 paragraphs and shrink them to a couple of sentences with lose most meaning. Having to hand hold it through each paragraph and then rewrite in my own style is never much of a time saver.