Field Notes

Why use local AI? Not just to save money

Today I stopped calling this a blog

Once the articles were written, I wanted a place for the work between them: small discoveries, unfinished experiments and the next thing to try.

The green terminal looked right. I couldn't find where to type.

Once Alex had the Matrix look, I began removing the borders, logs and notifications that got in the way.

Alex replied, but in the wrong conversation

Text chat could already retrieve information. A voice task went to another conversation, so I had to check where the work was landing.

Detect the source language. Keep the target I chose.

Automatic detection should save a repeated choice while leaving the translation direction in my hands.

The translation took seconds. Loading the model took longer.

Timing the stages separately showed where most of the wait was going.

I closed LM Studio. My monitor kept waking it up.

The tool checking whether LM Studio was running was also starting it.

Let the queue run. Let me move the urgent job.

Once several tools shared the local model, I wanted control over which job came next.

Finish one small task before buying the next piece of hardware

The laptop, enclosure and V100 came in stages. Trying the equipment I had helped me decide what to spend on next.

Open the part I use most

When even I couldn't tell which menu to click, I asked for transcription to be the opening page.

When it says done, I open the result

A transcript missing several minutes changed how I read progress bars. The same habit helped later with Alex.

The transcript was finished. Five minutes of speech were missing.

Listening to the recording again showed that the blank stretches contained speech the software had failed to transcribe.

I needed more room to fix the transcript

The settings filled most of the window. The transcript I needed to edit had only a few lines of space.