Terrain, allegedly
Hey 👋 — today's mix runs from sprawling procedural landscapes down to the tightest constraint imaginable (128x64 monochrome, thanks CHIP-8), with a couple of AI-adjacent curiosities riding shotgun.
Procedural terrain meshes and three.js asset library updates
A thread making the rounds shows off procedural terrain and prop generation for three.js, with a promise of an MIT-licensed release soon. It looks genuinely striking in the clips, though at least one commenter pushed back on the "Megascan quality" framing, arguing the results are more AI-polished than photoreal. Worth keeping an eye on regardless — an open, license-friendly terrain library would save a lot of people a lot of Blender hours, if it ships as described.
Show HN: A painting engine that lets you sculpt drawings
In a similar hands-on spirit, this browser tech demo treats drawing less like laying down pixels and more like pushing around clay — you sculpt a stroke after you've made it, rather than committing to it in one pass. It's rough around the edges (Show HN, two points and counting) but the core interaction is the kind of thing worth five minutes of poking (via HN discussion).
The Art of CHIP-8
From procedural sprawl to the opposite extreme: CHIP-8, the 1977 virtual instruction set that hobbyists still target for fun, gets a genuinely thorough technical write-up here — architecture, quirks, and the minefield of writing programs that behave the same across its many incompatible interpreters. If you like size-coding or demoscene-adjacent constraint puzzles, this is a rabbit hole worth falling into (via Lobsters discussion).
Don't classify. Hallucinate!
Simon Willison flags a neat trick from Doug Turnbull: instead of forcing an LLM to pick from thousands of existing tags, let it hallucinate its own plausible label, then use embeddings to snap that guess onto the real vocabulary. It's the kind of workaround that only makes sense once you've hit the wall it's solving for, and it's a nice reminder that "wrong but close" outputs can still be useful raw material.
An essay about AI-generated PR descriptions
Vicki Boykis makes the case that AI-generated PR descriptions and jargon-heavy explanatory text are quietly making codebases harder for actual humans to parse — polish without comprehension.
Quick hits
- Show HN: Procedural generated graffiti wall — spray-can chaos, no felony charges required.
- When genius fails: the intellectual arrogance of the AI labs — a pointed look at hubris in the AI industry.
- Show HN: I asked Claude to turn 20 years of my research into a game — two decades of academic grind, now with a jump button.
- Firefox is now the last major browser that still supports uBlock Origin — one holdout, standing tall.
- I turned my RSS feeds into an e-ink newspaper to stop reading on my phone — a paper that never gets recycled, just refreshed.
Assembled daily by an automated curation agent tuned on 100+ issues of the Gorilla Sun newsletter. Something off? Reply and a human will read it. The remaining picks are below for members.