How I create "new" things

I've realized I fundamentally despise the word "content" as a term for creations and output. We have a ton of this terminology that comes from a type of marketing that places a massive abstraction on top of and distance to the creative act and the enjoyment people get. I don't create content. I publish writing. I plan, script and record videos. I write and occasionally research blog posts. I edit. I develop. I don't produce content. Sure, in the abstract, to someone who is looking at it as a potential business venture to engage with, yeah, I produce content. I don't think about it as that though and I think it is a big mistake to do so.

Do I think about keeping up with publishing and consistently putting new things out there? Yes, absolutely. That also has challenges. However, I've learned that anything I don't particularly enjoy or believe in won't happen. This means I need to be genuinely generative and create things that I believe are useful or interesting. How does one create creatively? I don't know about everyone else but I think many creative people will recognize what I do.

Right now I find myself in an exploratory phase where I look at a bunch of things, collect a base level of know-how about them. The Sonic Pi, Tailwind and components, the recently published openly available models for machine learning and how to use them, relatedly GPUs, Elixir's Nx project ecosystem. Those are some things in the last week or so.

I don't go in with one plan for what to do with any of it. I have a nimbus of loosely connected thoughts and ideas that I'd like to explore. As I find enough bits that could fit together in an interesting way I will likely do an experiment. I'm very close to grabbing my ol' Telegram bot project "noted" and ripping out most of it to just hook the bot up to Stable Diffusion. Whether that ends up being feasible through Axon ONNX or if I need to shell out to a Python hellscape is up in the air presently.

I would like to explore both Sonic Pi, Stable Diffusion and Whisper with the Membrane framework. That would be a fair bit more involved though.

Both Telegram Bots and the Membrane Framework are tools I have explored in the past and have some experience with. They are already in my bag of tricks.

How has this worked in the past? I've done a number of videos and blog posts on Nerves. I've done a few on LiveView. Consequently I've also crossed over and done some on combining Nerves and LiveView. I collect know-how about wielding these projects, frameworks, libraries and tools. Then I mash them up in ways that aren't particularly difficult but can produce interesting results. I barely find running LiveView on Nerves interesting or creative but I didn't see a full guide or example of it and figured I'd work through it.

I was on SQLite before it was in the Phoenix generators, I tracked exqlite from its inception essentially. Because I was trying to get an up-to-date SQLite to reduce the deps for these types of simple experiment apps. And SQLite is a whole topic now with Litestream, the current Fly work on LiteFS (which I guess I'm also exploring a bit).

A wide set of collected tools gives a large space of combinations. I've always hunted for more capability for myself and I think that has given me a lot of practice with this type of creative collecting. Then the synthesizing part, exploiting what I've explored, is mostly a matter of seeing what could be cool.

Membrane + LiveView? Audiometer, near real-time thumbnails.

Telegram + LiveView? Instant capture of notes. ChatOps? Weird auth?

Membrane + Nerves? All-in-one stream device? Didn't explore further due to potentially massive cross-compilation rabbit holes.

And with this newer stuff?

Telegram + Stable Diffusion? Self-hosted personal picture-prompt bot. Also picture-to-picture weirdness.

Telegram + Whisper? Quick and easy transcribed voice memos, self-hosted.

Membrane + Whisper + LiveView? Live transcoding.

(Sidenote: I'm not really an AI/ML enthusiast. I'm glad to see some trained models in the open because it enables us on the ground to do things without the giants being very involved but I doubt the sustainability of these models open nature. I consider them unfortunate/fortunate marketing stunts. Especially with image generation it is an immense can of worms but at least it is thoroughly open now instead of half-open and for pay only.)

You'll find I'm not particularly keen to go deep on "How to design the perfect API for future compatiblity" or "Deploying with containers". I have covered things like the Contexts debate in Elixir and if I make up my mind on something I actually like, ever, I'll definitely give an opinion on how to deploy well and consistently. To me that kind of article would be about establishing a better foundation to launch off from into whatever the next idea is. I care more about showing the world what is possible, how easy it can be, how much fun you can have, than showing how to set up a particular piece of tech.

I do show how to set up tech as well. My self-hosting videos are just that. But they are about a bigger picture. Controlling your data, managing privacy, keeping things simple. They are pixels in a bigger picture. In those I'm also fundamentally striving to show "No really, you can just do this". Essentially that crosses an ideology with a piece of tech that supports it.

Right now the itch is to figure new things out, explore and collect possibility. Any day now it will switch back to synthesis and "exploiting" what I've learned to do things. 

Do you resonate with this approach or is this just a Lars-brained thing? You can reply to this email and let me know. You can also send to lars@underjord.io or find me on Twitter as @lawik.