What's this about?
A developer built a voice companion that plays Skyrim alongside them in real time, with the engineering focus on keeping response latency low enough to feel like a person rather than a chatbot.
What exactly is "latency" here, and why does it dominate?
Latency is the delay between you speaking and the companion answering. For most AI uses a two-second wait is fine. In a game it is fatal — the moment has passed. So the entire build is shaped around shaving delay: smaller models, streaming speech, starting to answer before the thought is complete.
What's the community saying?
HN 389 points, 76 comments
- Appreciation for a build where the constraint is latency rather than capability, which produces very different engineering decisions.
- Interest in this as a preview of NPCs generally, which connects to the DeepMind games partnership below.