Part one — media
Flat media is a historical accident. AI removed the content barrier and glasses-free displays removed the hardware one. The reason it will stick is older than either: depth feels more real, and what it does best is presence. A face you can meet. A memory you can stand inside.
- 0113 June 2025
The Rise of Spatial Content: A New Norm in Digital Experiences
Apple's WWDC confirms 2D-to-3D conversion is going mainstream — and quality hinges on depth models trained on real stereo, not synthetic data alone.
3 min read - 0217 July 2025
Immersive Media and the Power to Enhance Human Connections
Immersion displaces flat media because it feels more real, and its highest use is presence — which is why the breakthrough is headset-free, eye-tracked systems.
3 min read - 0319 August 2025
Immersive Memories – Fragments of Time Relived through AI
The photogrammetry pipeline behind Black Mirror's "Eulogy" is being commoditised by consumer 3D capture, video diffusion and Gaussian splats. A human development, not just a technical one.
2 min read
Part two — machines
The stakes change. 3D stops being a better way to watch and becomes how machines learn what is real. Appearance can be generated, geometry has to be measured, and these essays keep returning to where measured observation comes from and who checks it. Plausible is not true. The endpoint is still human: safer machines, and memories that can be rebuilt without being invented.
- 0416 September 2025
Building the Foundation for Spatial Intelligence
The next leap in AI needs internal 3D world models, but no web-scale 3D corpus exists — and video entangles camera motion with scene dynamics.
6 min read - 0521 October 2025
Calibrating Spatial Intelligence
From a single eye to a physically accurate world
Self-calibration and monocular metric depth can go a long way on one eye — but scale stays ambiguous without real geometric constraint.
6 min read - 063 December 2025
From Meshes to Splats
The new language of spatial AI
Meshes are workable illusions, radiance fields are realistic and slow, and Gaussian splats are the first medium expressive enough for an AI to externalise its world model.
5 min read - 0720 January 2026
The Geometry of Common Sense
A humanoid robot walking into a mirror is a world-model failure. Passive video-trained models mistake appearance for reality — and are structurally compelled to hallucinate rather than admit uncertainty.
5 min read - 081 April 2026
The Missing Interface of Spatial Computing
Why spatial displays need their OpenXR moment
Spatial displays have shipped from Samsung, Sony, Acer, Lenovo and Looking Glass, but most still drive them through proprietary SDKs. What OpenXR would need in order to describe how these screens actually work.
6 min read
Written in a personal capacity. David Fattal is founder and CTO of Leia Inc., and founder of the DisplayXR project.