as expected from mappletons.
i love reading her garden!
this tool ↓ is obvious in hindsight, and u can apply it across different areas of your life. my pivot away from AI companion actually stemmed from this. whether it'll work out is still TBD.
looks like you can export the splat and load it using Spark on the web or UnityGaussianSplatting plugin in unity then drop your VRM on top via three-vrm or UniVRM.
so.. the world-building part for AI characters is solved?
We’re opening up access to Atlas to everyone today.
We’ve built apps and API endpoints to help you use Atlas any way you can imagine: bring your ideas for building a 3D world and come away with keyframes, video clips, or 3D Gaussian splats.
once a haafu (half-waifu, half-assistant) enters the game, it's owarida.
i don't think the ratio even matters. just look at tolan, i think it has a good balance of assistant capabilities that's more than enough for general consumers.
imho that's where things headed.
no overpromising. no full automation.
just like engelbart envisioned (ok, maybe i was stretching it a bit)
saw 3-4 new startups building voice eval/benchmark sites lately.
are they trying to be the Arena or AA for voice? none of their results line up with each other tho tf
a new startup drops a full-duplex model every few days, literally.
just waiting on a quantum-isomorphic triplex at this point, o/w i'm not gonna bother!
confession: i miss talking to a MEAT coworker.
talking to a coding agent cut my lifespan in half.
seriously, i need a "waifu" layer to shield me from all the stress these agents put me through.
i feel like i'm becoming a worse person just by talking to them, which is the exact opposite of how talking to an AI companion feels.
it's the difference between paying for something you want vs something you're trying to avoid. not that i hate coding, i just suck at it compared to the frontier models.
security is tough, man.
personally, i don't really care about ZRD cuz i know my system inside and out and know what's safe to share and what isn't.
besides, i actually want the model (whale) to get better at my specific use cases, so take my data! tbh its EQ still isn't great, though maybe that's just a skill issue on my end.
when i'm shipping a production app, proper ZDR across the entire stack is a non-negotiable. it's not cheap eithter, and it's almost always locked behind most expensive tier or a "talk to sales" wall.
@RealRichomie thx for open sourcing this!
i noticed at 0:26 and 0:40, awaken's voice kinda faded out?
also, random side note: it actually sounded more natural than the "user" lol