@ZainHasan6 — the clearest explanation of LLMs as world models I've seen
The most clearest and crisp explanation, I've ever heard, of how large language models compress and capture a "world-model" in their weights simply by learning to predict the next word accurately.
Furthermore, how the raw power of these base models can then be tamed by teaching…
That framing — next-token prediction as implicit world-model compression — cuts through a lot of the usual LLM mystery. Most explanations start from "stochastic parrot" or "just autocomplete," both of which miss why these things are actually useful. The second part, about taming base models through alignment, is the part I wanted to read more of (the tweet cuts off into a video). Still, the setup alone is worth saving.
@pissang1 — web app that converts SVG to 3D with ray-tracing
🥳🥳Finally I published the web app version of "Vector to 3D": https://meimu.design/vector-to-3d/
Like the Figma plugin, it converts the uploaded SVG to 3D with a high quality ray-tracing render. It can be used to create 3D icons, text, illustrations, or other 3D-style images with simple…
Smart, because ray-tracing is usually the part that requires a full 3D pipeline and hours of render time — doing it in-browser on SVG input is genuinely clever. The workflow for creating those slick isometric-style icons just got a lot shorter. Bookmarked to try on some of the flat icons I've been wanting to give depth.