On my mind this weekend:
- I think the distribution channels for commercial software are broken.
What's funny is that AI should make this easier, actually. But at least right now... nobody's really thinking in terms of building out new channels, or thinking about new distribution models and licensing types for software.
I'm thinking that market segmentation is a good thing, actually. And that the bootstrap problem isn't as serious as people usually think it is.
I'm investigating the feasibility of building a large scale collaborative project where we can all sit and brainstorm new models for distribution channels. This makes more sense if we fan out and attack it as a larger group.
But collective support and an organized collaborative project is probably necessary.
My thinking is that if the community could put out 11,000 channels... and 6 of them provide high quality community building or software distribution infrastructure that's worth using, everybody wins, and it doesn't even matter who ends up owning the thing.
The trick is human intention.
As usual.
Painting controls
- Generative classifiers for ultra-lightweight transcription based control systems and simple code generation tasks. I'm writing up a framework to make this a little easier, using RNN's. You don't need transformers to do this kind of thing.
It would solve some of the problems I've had trying to build Kinetic Painter so far. That's the one where you embed paintbrushes on your fingers and tongue and follow them with motion capture over a gaming engine that knows how to make the paint dry.
My first thought was that I wanted to build the canvas first, but the canvas is replaceable and incidental. Really, the control system is what makes an app like that special. So it actually does make sense that this is the layer with all of the performance bottlenecks and technical problems.
Anyway, if you like to play with RNN's and light model trainers, I'll have something on GitHub posted this weekend. You can give it a spin.
Reconstructing older software
- The Flash5/Codecharge contract study is going quite well. I think my point of nostalgia with Flash is the fluidity of the thing. And I'm almost at parity with the base, before extending it into something imminently ridiculous.
The codecharge piece is ready for base testing, and I hope the stuff it spits out is at least somewhat nicer than the original. I think the regeneration problem is especially interesting.
Other experiments and Petrolina
- Ruby Tumbler is a rails compiler I'm working on. Need to rest. It's making my poor little brain hurt.
- Recursive model training project. Currently on Corpus 6. There's been progress, but nothing as dramatic or interesting as I was hoping for. Maybe a little bored because all of my recent hypotheses are failing. I'm taking a week off on it. When I pick it up again, I'll evaluate if I think I can get it done now, or whether we should probably wait until Qwen4.
- Petrolina has a new donor truck and is ready to have her brain put in Mwhahaha.