Productivity/collaboration catching up to Ventrilo + CS UX!

RT if you want day-zero support for OpenAI new open-weights model to run fast on your laptop with MLX.

wave through sediments of ideas: made a conceptual interface of using hand to traverse layers of commentaries on plato’s timaeus, from aristotle, to christian theology, and to today

1/2 the team that animated Flow (won the Oscar for best animated feature) had never used Blender and 1/2 were junior!

found this old demo that uses local AI to dive endlessly deeper into a concept

locallama buying Chinese modded 4090s for more VRAM from vendors that allow you to send in hw for modding is something out of a cyberpunk flick

Manga panels for vision models benchmarks

Some interesting bits in the linked posts on building on-device models to automate phone actions: - Minimal difference in accuracy training on text based UI trees vs. image screenshots, but latter was more efficient using fewer tokens and had more room to learn w/o overfitting -

Experimenting with local screen recording analysis in the background with FastVLM + summary with Claude.

AI services will be increasingly hungry for our personal context. We dug into the dynamics here and explored how the future might unfold.