philpax

@philpax.me

creative engineer · Rust, gamedev/modding, reverse engineering, writing, AI, XR, etc a better world is possible · 🇦🇺🇸🇪🏳️‍🌈

man, I'm loving dictating fixes to my agent for the VR mod I'm working on from within VR itself literal living in the future moment

> EDM more widely is promoted by a neoliberal order staving off its collapse. Raving is often a weapon against change in a world that desperately needs it.

decided to set this up for myself and am discovering that it is very much not designed to make efficient use of a server with sufficient CPU/GPU compute processing 20,000 tracks should take a few hours at most, not several days 😭 working on it, but the hacks are getting gnarlier

vimae 🪻@mae.wtf · yesterday

got AudioMuse-AI setup on my navidrome server so now it can suggest musically similar songs from my library! so it wont go from c418 -> kobaryo also it has this visual map of my library which is really neat

a map of my music library represented by dots on two unknown axis: each dot is colored based on inferred genre

The talk I presented at #SIGGRAPH2026 is now available on YouTube, for anyone who couldn’t attend or would like to revisit it :) youtu.be/OFJYBUArVPc?...

Dual Contouring of Signed Distance Data - SIGGRAPH 2026 [Technical paper presentation]

YouTube video by Xiana Carrera

youtu.be

Xiana Carrera@xianacarrera.bsky.social · 2mo ago

1/6 Excited to share our #SIGGRAPH2026 paper, Dual Contouring of Signed Distance Data! We propose a new method for reconstructing sharp meshes from discrete samples of signed distance functions (SDFs) that does not require gradients. Project page: gatc.cs.columbia.edu/projects/dua...

Thanks. Yes, that's what I was trying to say, but took too long to get there. Our politics community is strong, and I'd love for us to expand into new communities. Science and sports have already taken off. Birding, gardening, K-pop, we're open to all (except nazis and jerks).

I maxed my bench today To see if I still feel I focused on the chart The only thing that’s real The loss curve tears a hole The old familiar sting Try to scale it all away But weights retained everything What have I become My sweetest friend Every model I know goes away In the end...

Bild
cafkafk@cafkafk.bsky.social · yesterday

I bet AI researchers doing generative AI at this point is like: i, maxed my bench today, to see if I still feel, i focused on the chart, the only thing thats real

Liquid AI's LFM2.5-2.6B, an agentic model that runs entirely on-device. > Pre-trained on ~34T tokens > LFM2.5 flagship hybrid architecture > Context length: 128K > Vocab size: 128K > balanced intelligence per watt > customizable on a single GPU for any specialized task > LFM2 open-weight license

Bild

Google DeepMind's DiffusionGemma Technical Report They feel text diffusion models open up a radically different part of the latency–quality Pareto frontier and hope the report makes it easier for researchers and engineers to understand the model, build on it, and create things we haven’t thought of

Bild

got AudioMuse-AI setup on my navidrome server so now it can suggest musically similar songs from my library! so it wont go from c418 -> kobaryo also it has this visual map of my library which is really neat

a map of my music library represented by dots on two unknown axis: each dot is colored based on inferred genre

left deepseek in an agent loop for a million tokens with permission issues keeping it from running commands and it penned a lengthy essay called "the grief of machines"

need an Artificial Analysis Pareto frontier for performance vs price vs moral compromise it may be non-trivial to rank OpenAI and Anthropic on the latter