How Did We Get Here?
Why closed-source sucks
There's always a chance someone could be lying about a service if you can't prove it yourself. Tools like ChatGPT can claim security and privacy. Sure, you could put trust into the company, but you never know what you're buying. With open-source models like Qwen and DeepSeek, you can inspect what's happening behind the scenes.
What started my love for open-source
My first appreciation for open-source probably came with my Raspberry Pi. Raspberry Pi OS (formerly known as Raspbian, and I wish it still was) is almost fully open. When I got it, I started finding what open-source was and how it worked. And, honestly, I still don't know how I got here. All I know is that I did.
My switch to open-source
Okay, I'll admit it - I used to use ChatGPT.
HUGE COMMUNITY GASP
My transition to DeepSeek was basically a little integration between the two. I had ChatGPT chatting with DeepSeek, and i thought it was really cool. Now I think ChatGPT sucks.
Additionally, I did switch to Claude for a little bit, but only because DeepSeek was blocked at my public library. (I hate c********m.) The rate limits were killing me.
How I ended up making my own
I really love old electronics. Hell, I'm even writing this on a 2016 laptop. And - for some reason - floppy disks really intrigued me. So, I aspired to make a model the size of one.
Behold, the Single Floppy model!!
And, somehow, it scores worse than a 1k model. Nice.
...one day, it all blew up.
AND SO I MADE...
MORE!!!
floppyx3 was terrible. The tokenizer just failed... somehow. floppyx4 I never finished. And now I'm thinking about buying a whole damn PC just for training.
TL;DR
Don't.
