El “portal” de Vitre mezcla captura 3D, Atlas/Marble 2 y perspectiva ajustada al observador. La demo se ve increíble, pero su autor aclara un límite: convence más desde una cámara que al mirarlo con ambos ojos. https://t.co/xX39YhxHfr
Vitre Experiment #6… my favorite one so far!
Same hallway, new portal effect opening onto a completely different world.
Here again, I used Atlas and Marble 2 from @theworldlabs to reimagine the hallway from my original 3D capture. I’m having a lot of fun exploring what these tools can do, and I’m really impressed with the results.
As with my previous experiments, the idea behind Vitre is to make the monitor disappear by aligning a virtual scene with the physical world. And of course… everything is 3DGS, rendered in real time.
#3DGS #WorldModels #GaussianSplatting
Haciendo pruebas con #threejs terminé construyendo esto: un simulador de artillero AC-130 que corre en el navegador, también en móvil.
Torreta EO/IR, cañones de 30 y 105 mm, oleadas enemigas y cuidado con el fuego amigo.
Pruébalo 👇
https://t.co/g7vlKqe4m3
#gamedev#webgl
@JohnGreenDev Fair point. If the spec is wrong, green tests can just reinforce the wrong behavior. I would want acceptance examples reviewed independently of the generated code, with human review focused on ambiguous requirements and high-impact changes.
Antes de cambiar de modelo, revisaría cuántas llamadas del agente repiten decisiones ya tomadas. HarnessMM explora reutilizarlas bajo condiciones explícitas. Me interesa medir el ahorro junto con las regresiones, no solo contar tokens. https://t.co/3NcML1qqLz
@omarsar0 The guard design seems crucial here. I would want to measure token savings alongside regressions on tasks where tool state changes, rather than treating a cached decision as valid just because the next action looks familiar.
@laravelnews@ericlbarnes Starting with an existing app that has almost no tests is a useful angle. Does the course cover choosing which critical flows to protect first, and avoiding tests that break on harmless refactors?
@carlesnunez Lo de reutilizarlo como panel de datos me parece lo más interesante. ¿Una vez configurado puede funcionar totalmente en local, sin depender de la nube del fabricante? Ahí le veo mucho valor para darle una segunda vida al hardware.
@thesaraharminta Local + private is the part that caught my attention. How will export work: can users keep their notes and structured data in an open format, including relationships and formulas?
@coolerwave Worth flagging for anyone trying this: you're giving an agent read access to 3 years of private media. Run it on a copy, offline, and check what ends up in logs and transcripts. The cache finding is real, but the audit itself becomes a new exposure.
@RoundtableSpace OpenShell is the one I'd look at first. Kernel-level scoping plus agents never touching real credentials fixes the biggest gap I see in agent setups today: a full .env mounted into the container "just to make it work".
@hqmank Matches what I've seen: most handoffs I built for older models were just context loss with extra latency. Where subagents still pay off for me is isolation: parallel work in separate git worktrees, or noisy exploration I don't want polluting the main context.
@Bagdex Code stopped being the moat a while ago. What people still pay for is someone else running it: uptime, backups, upgrades, retention. Self-hosting analytics is free until the DB fills the disk at 3am.
Now on Linux! I ported the screenshot clothesline to GNOME Shell 49.
macOS: https://t.co/SzxG2D7TSr
Windows: https://t.co/kVBJaX4Z4m
Linux: https://t.co/tFazC4TQ0v