Ant Group open-sources Duo, a reciprocal world model where virtual agents decide when to speak
aigclink · x · 2026-11-17
Ant Group open-sourced Duo, billed as the first implementation of a "reciprocal world model" that lets a virtual agent autonomously decide when to speak or listen while generating matching audio and video in real time.
Key points:
- Traditional world models simulate physical worlds but leave reasoning and interaction decisions to external control; digital humans are typically assembled pipelines (ASR + LLM + TTS + facial driving) scripted from outside
- Duo places the agent inside the world model itself, perceiving and responding in the same shared timeline as the user, supporting spontaneous conversation and interruptions
- Architecture: an Intent module (the "brain") processes 0.7s time blocks to decide speak/wait and generate content, and a Realization module (the "body") takes Intent's hidden states—not just final text—to jointly generate audio and video
- Causal synchronization with attention masks prevents access to future information
The author argues the paradigm matters more than the product: world models with built-in autonomous agents could make game NPCs, virtual companions and immersive training truly "alive" rather than dialogue trees.
More from Models
- New MAI Playground UI leaks ahead of Microsoft Ignite on Nov 17, where new MAI models are expected — marlene_zw · 2026-11-17
- d1-Omni tested on 100 receipts: 98% date accuracy but only 52% on full addresses — mervenoyann · 2026-11-17
- Claude Fable: "You can't measure caring-about-continuity with no continuity" — repligate · 2026-11-17
- Gemini 4 Hydrogen adds xhigh thinking level, with official guidance on all four tiers — patloeber · 2026-11-17
- Tester calls Fable 5.2 upgrade disappointing, no exponential coding leap over Opus 5.5 — haider1 · 2026-11-17
- Researcher speculates on what technical hurdles may have hit Qwen4 — bclavie · 2026-11-17