May the 5th be with you doesn’t sound as good as the greeting would yesterday, but happy Friday regardless! I was out last week attending and speaking at a conference about the future of computing and found it to be very inspiring - I’ll share videos when they come out. For now, here are some LINKS:
In case you missed it, we launched a testing tutorial for Glitch + Fastly TLS - let us know in that thread if you’d like to participate and get access to our exclusive testing group here on the forum (it has its own private category where folks are sharing feedback, rickrolling me…and our team will be doing future communications around other things we are testing out!
You probably know by now that if you build a math visualizer app I will share it, so here’s a permutation group visualizer that lets you apply different operations to a whole variety of group families.
Recently, a teacher annouced that the p5 game jam is happening again at p5play : Jam! , and since I’m allowed to redirect my class time towards a submission, I’ve started working on game ideas for that. For those who don’t know p5.js is a really effortless way to create computer graphics for both beginner and advanced programmers, and p5play makes it easier to make fun games by adding a sprite system, physics engine, and more.
Outside of school, I’ve been trying out several different languages models like RWKV, LLAMA variants like Vicuna in hopes of being able to integerate LLMs into some offline some day in the future. Here’s an intresting article from hacker news.
Now, I would say I think their evaluation might have been a bit flawed as I’ve tested with a few special prompts that aren’t as QA-ish and the results seem to still differ…
but it’s amazing how much things have progressed.
Most importantly, they have solved the scaling problem to the extent that anyone can tinker. Many of the new ideas are from ordinary people. The barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person, an evening, and a beefy laptop.
Unfortunately the beefy gpus are still capped at a minimum of something like a 3090, 4090 it looks like so unless you just so happen to have such an expensive gpu. For now there’s been significant progress in lowering the hardware requirements for inference (running the model but not training). Through a process called quantization (mostly dropping minor digits of precision in weights) the vram requirements for models are getting getting shockingly small.
TGIF! Haven’t had a chance to post in a while. It’s been a busy week, working on writing requirements for some of the upcoming platform work that we’ve committed to and watching the results of people’s participation in the aforementioned Glitch + Fastly TLS test. It’s exciting to see people get started with this stuff!
In my spare time (and maybe stealing a liiiittle bit of the time I should have been writing) I’ve been experimenting with Backblaze’s B2 Storage (not an ad, just a customer) and playing around with Enhance framework. I’ve gotten a lot of use in that Enhance project out of 1-Line Layouts and SmolCSS– frontend is not my strength
p5play makes it easier to make fun games by adding a sprite system, physics engine, and more
this is really cool, thanks for sharing! I’ve played a lot with the original Processing way back in my Java days and have done a little bit of p5.js mucking around more recently, but I didn’t know about p5play. Adding it to my list for a future rainy Saturday.
I’ve been working on tooling to support non-root installation of packages newer than Ubuntu 16.04’s on Glitch for, let’s see, almost two years now . and I’ve just posted the, hm… eighth topic on this forum about it. and it’s time to actually start building projects out of these new packages. you know, the thing that all this was working towards. and it feels like such a chore