---
schema: 1
type: term
title: '**Mirage**'
description: >-
  **Mirage** — a model architecture referenced in Decart's real-time
  video-to-video AI work as explored in this channel's conversation with CEO
  Dean Leitersdorf. It appears alongside the broader shift toward **newer AI
  architectures and form factors** — energy-based models, hybrid
  diffusion–autoregressive systems — designed to solve problems where LLMs hit
  their limits, such as **spatial reasoning**, **exact determinism**, and
  **real-time video**.
canonical: 'https://liquidgold.cc/channel/everyinc/explore/term/mirage-da2530e4b02c'
markdown: 'https://liquidgold.cc/channel/everyinc/explore/term/mirage-da2530e4b02c.md'
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*Mirage\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/mirage-da2530e4b02c

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**Mirage** — a model architecture referenced in Decart's real-time video-to-video AI work as explored in this channel's conversation with CEO Dean Leitersdorf. It appears alongside the broader shift toward **newer AI architectures and form factors** — energy-based models, hybrid diffusion–autoregressive systems — designed to solve problems where LLMs hit their limits, such as **spatial reasoning**, **exact determinism**, and **real-time video**.

## Usage in this channel's material ### Where it appears The term **Mirage** surfaces in the conversation about _real-time video-to-video AI_ with Dean Leitersdorf, co-founder and CEO of Decart[^08615b13-5661-5b2b-bcb0-3e4b551b9b63@126]. Within the topic **Beyond LLMs: New AI Architectures and Capabilities**, it belongs to the same thread of discussion as the newer model architectures — energy-based models and hybrid diffusion–autoregressive approaches — built to handle problems that push past LLM limits, such as spatial reasoning and exact determinism. ### What the video explores The episode covers: - how **real-time video-to-video AI** works - why it required **GPU assembly** and a **new model architecture** to build - what it implies for **gaming, creativity, and the future of work** ### Footnotes [^08615b13-5661-5b2b-bcb0-3e4b551b9b63@126]: This AI Makes a Video Game World in 40 Milliseconds.