part of me feels that forcing Yumon to be a coding model could be a rushed conclusion. maybe he is better off as a companion. he has never been as enjoyable to talk to as he is now
6
everyone is like, bro, go smol. im already tiny. maybe i stay tiny? i bet i can squeeze more performance out of 27M params. Even the smallest frontier models tend be hundreds of millions of params, so at least 10x, up to 30x. you don't truly need that capacity in theory
11
I can't say that anything will get me GitHub stars, but so long as my audience is engaging, I can collect funds and distribute products. You really can't force the big successes. You just need to let them burst on their own.
11
True Error is a great game. Yumon is smarter than ever. Indie Machine is stronger than ever. No cash yet. But I can wait! I've got it all set up now, Buy Me a Coffee on each post. So it's just a matter of time! Everything I offer is free so I rely on donations.
10
Some more volume is necessary to have possibility of earning enough revenue, just as a side business. However, not monstrous sizes, at the risk of losing audience quality. So what can I do besides be patient? I just don't know.
1
21
I could have chased trending topics, which makes some sense, but I discovered that my audience is higher quality. So why give that up? Courses fit the slow roll too. But how do I still achieve at least 10x volume over the long run? i don't know. i just hope it takes consistency
10
if I make Indie Machine into an ML eLearning Platform, maybe I will make a Tensor Crunching Library called Hyperfield. goal is lightweight, simple, easy visualizations, a wide variety of built in components. easy web-based training and inference.
11
imagine if Indie Machine became an LMS for ML Courses exclusively, as well as the content branch on the side. I have many deep, niche topics in ML to cover. Can create resources you'd pay for, Table of Contents that is compelling. Interactive Courses.
15
curiously curious about this or Patreon hmmm what is the right next step for Indie Machine to help it grow and generate revenue buymeacoffee.com/
15
complete satisfaction, finally. yet i am defiling it by scaling it further just to see what happens. the real goal is a viable coding model but i needed to prove this first
17
i got me PyTorch on the training runs with Rust Burn still handling inference and deployment (often baked into apps). quite ideal for Runpod runs especially
1
18






