2025-01-01

Last modified: 2025-01-01 20:02:34

• Clock
• LLM assistant

Clock

I'm going to put the remontoire driving side on with just 1 bearing, just to see if it will work. I'm not 100% sure there is enough room. If it works then I can think about making an alternative bearing block with 2 bearings in it.

It looks to be almost working. I'm driving it by applying pressure to the remontoire driving side with my finger.

Issues are:

• the balance pivots are only located by the thread, so every time you adjust them vertically, they also move around horizontally
• the remontoire weight has to be shorter than before to fit inside the frame, so it isn't heavy enough now
• the remontoire driving side bearing is too wobbly
• the remontoire catching arm seems to get caught at the wrong position, not sure why

Solutions:

• maybe redesign the balance pivots so there is a cylindrical feature for locating horizontally, and the thread only moves up/down?
• make the remontoire weight heavier
• make a proper bearing for the remontoire driving side
• have a look in CAD and work out why the remontoire arm angle is wrong

Forget the balance pivots for now, revisit that later.

I used the old winding drum as a remontoire weight, and it now looks like it runs! But amplitude only about 145 degrees.

So next thing is CAD for the bearing. Ideally it would reach over the roller lantern pinion in a U-shape to support the shaft closer to the middle. Not sure if I have enough room for that:

I think just make a slightly shorter version of the double-ended bearing that supports the escape wheel. And maybe come back to it in the future and make everything a bit more delicate.

Seems to be working OK! And feels much more legit than expected. Amplitude is about 160 degrees now.

So next up is putting the winding drum on and actually letting the clock run.

LLM assistant

The idea is that I'll have some continuous voice chat open with ChatGPT.

I'd want a "push to talk" input method. So it records a message while I'm holding my PTT button, and sends it when I release it.

The first pass is voice chat in OpenAI Playground.

It would be cool if it also got a screenshot of my PC, or photograph of my environment, from the time I first pressed the PTT button.

I want to have it open all day and just periodically converse with it.

I think one thing I'm struggling with is that I don't find it natural to just speak out loud. I think switching to ChatGPT and typing to it is more comfortable.

The voice chat in the OpenAI Playground is a lot worse than the one in the ChatGPT app.

We can imagine a headset with:

• bluetooth headphones
• bluetooth microphone
• webcam
• some sort of button

The button is "push to talk", and when you press it it takes a picture with the webcam and starts recording audio, then gets gpt4o to respond.

Instead of headphones you could use some AR glasses or something, but for text it seems like not much benefit over headphones.