Research

Fleek's efficient inference and model optimization technology is a result of the R&D done by Weyl, our internal AI lab. This page provides an overview of Weyl's research, papers, and the technical foundations powering Fleek.


Research Papers

Foundational research from the Weyl Archive, preserved in their original institutional format.

A Fistful of Errors

The Lattice Hypothesis — Proposing that neural networks fundamentally operate on discrete floating-point lattices rather than continuous functions on ℝⁿ. Traditional continuous analysis is reconceptualized as the approximation, with the discrete lattice structure being the computational reality.

Read Paper →

Curiouser and Curiouser

The Hallway Hypothesis — Examining how constraints that reduce wrong moves matter more than constraints that reduce total moves. Analysis of phenomena like LoRA effectiveness and quantization failure modes reveals the geometry of transformation corridors in neural network pipelines.

Read Paper →

Not Even Long

The Landauer Hypothesis — Treating precision not as a hyperparameter to optimize but as a physical quantity to measure. Explores how the thermodynamic cost of bit erasure constrains low-precision quantization schemes, with implications for NVFP4 deployment at the edge.

Read Paper →


Technical Articles

Deep-dive articles from Weyl .plan, covering CUDA architecture, tensor cores, and optimization techniques.

The Villa Straylight Papers

A comprehensive series on CUDA architecture and tensor cores, encoding NVIDIA's theorems as types through Gibson's lens.

PartTitleFocus
OverviewThe Villa Straylight PapersJensen's Razor and the malevolent combinatorics of CUDA architecture
Part IThe Rectilinear ChamberLayouts, Coordinate Spaces, and the CuTe Contract
Part IIThe Sense/Net PyramidCoalescence, Noetherian Reduction, and termination proofs
Part IIIBuilt Him up From NothingComplementation, FTTC, and iteration space holes
Part IVTake Your Word, ThiefComposition and the Tensor Core Cathedral

The Inhuman Quality of Starlight

The Operating System of the Drone War and The UTF-8 of AI — Part 1 explores how constraints dominate resources, and the lattice doesn't negotiate. Covers NVFP4, infrastructure, DeepSeek, CUDA, Nix, embedded AI, and quantization.

Read Article →


Benchmark Methodology

For details on how we measure and report performance in our benchmarks:

Benchmark Methodology →


Stay Updated