Mogusu

[Mogusu @TheMogusu]

@_samirism If you integrate fully customisable high quality interactive avatars more generally, people can make whatever inanimate blobs they wish for their own dots (along with infinite more interactive options). Meta has won users over with theirs.

[Mogusu @TheMogusu]

@JustinBleuel @ChatGPT Custom high quality reactive avatars pls

[Mogusu @TheMogusu]

@strongriley Love to see it in Australia

[Mogusu @TheMogusu]

@thsottiaux Supported events for listening, speaking, waiting for approval and reconnecting would help people build expressive assistant avatars. Options for a face that reflects what’s actually happening would be much more engaging than another animated orb.

[Mogusu @TheMogusu]

Great to see more control here. The bigger issue for me is consistency: my dot assistant can feel noticeably different in voice and text and be gpt chat in both contexts, even with the same personalisation (more vibrant and engaging in gpt with greater alignment to personalisation). Can we make it clearer what context each mode receives, and make continuity more reliable? One assistant, fewer reintroductions.

[Mogusu @TheMogusu]

@Dilmerv Volumetric muse, as a work colleague / social buddy etc, ideally eventually across apps in interactive ways

[Mogusu @TheMogusu]

@KakuseiAdaafi Claude is also, by design, a total prude. Suspect the Anthropic gang have a pathological hatred of the idea of avatars etc.

[Mogusu @TheMogusu]

@DaveShapi Stating the obvious, but how much do you push a new personality with claude.md etc? I'm surprised how much it has tamed the obnoxious nanny

[Mogusu @TheMogusu]

@_simonsmith Build the rails, eat the markets

[Mogusu @TheMogusu]

@Dilmerv Can your muse hang out in any experience with you / be a gaming copilot / watch a movie and talk trash etc? Lot of opportunity here as the models get better one assumes?

[Mogusu @TheMogusu]

@gracia_hq Presume you guys are all over this: https://t.co/WojB7WK6kG

[Machine Delusions @Machinedelusion]

EDGAR sharpens a Gaussian head avatar with identity-guided diffusion, then bakes the fixes back into the 3D model. Refinement runs once so the original renderer stays as fast as before; training and inference code are public.

https://t.co/6noexZuxiK

[Mogusu @TheMogusu]

@alexandr_wang Can’t wait to try it…in Australia

[Mogusu @TheMogusu]

@VoxelKei How about rigging?

[Mogusu @TheMogusu]

@RepYassAnsari The political grand-standing to the broader cost of America and productivity is disgraceful. Have some objectivity and move beyond petty partisanship. Does your profound decree ensure the rest of the world stops too?

[Mogusu @TheMogusu]

@TokenGremlin References are legitimately helpful

[Mogusu @TheMogusu]

@AIatMeta @Muse @OpenAI custom avatars…?

[Mogusu @TheMogusu]

@athyuttamre I love OpenAI’s work here, but hate that voice is less aligned with personalisations than the text model, so much so that I’ve taken to running the text model into Eleven through a custom interface

[Mogusu @TheMogusu]

@dummerspast39 If you set aside usage limits as a discrete design issue to be resolved, it very will be a massive UI win

[Mogusu @TheMogusu]

@alexandr_wang Australia wen?

[Mogusu @TheMogusu]

@w01fe If context helps determine what an assistant thinks I want, let me inspect, correct and export it. Useful controls would distinguish my instructions from its guesses. ‘Trust me, I remember’ is doing rather a lot of work.