ART
&
TECH
Jason Lee

JASON LEE

My vision is to bridge art and technology.

We're at the edge of a new realm of entertainment, built not just on LLMs but on deeper fields like computer vision and reinforcement learning.

The future of creative experience lies in systems that perceive, adapt, and respond, not just generate.

Art Practice

2023 — 2025
Wastelands 2

Wastelands 2

Acrylic on canvas · 2′ × 3′ · 2024
Wastelands 3

Wastelands 3

Acrylic on canvas · 2′ × 3′ · 2024
Wastelands 4

Wastelands 4

Acrylic on wood, silkscreen Lokta paper · 18″ × 12″ · 2024
Wastelands 5

Wastelands 5

Acrylic on wood · 18″ × 12″ · 2024
Sultry 1

Sultry 1

Acrylic on canvas, collage · 18″ × 12″ · 2023
Sultry 2

Sultry 2

Acrylic on canvas, collage · 18″ × 12″ · 2023
Sultry 3

Sultry 3

Acrylic on canvas, collage · 18″ × 12″ · 2023
Sultry 4

Sultry 4

Acrylic on canvas, collage · 18″ × 12″ · 2023
2020 — 2023
Pour

Pour

Gouache on wood · 2′ × 4′ · 2022
Cyanotype 1

Cyanotype 1

Cyanotype on watercolor paper · 7″ × 18″ · 2021
Cyanotype 2

Cyanotype 2

Cyanotype on watercolor paper · 7″ × 18″ · 2021
Cyanotype 3

Cyanotype 3

Cyanotype on watercolor paper · 7″ × 18″ · 2021
Hide 1

Hide 1

Hand-made pigments on paper · 8″ × 11″ · 2021
Hide 2

Hide 2

Hand-made pigments on paper · 8″ × 11″ · 2021
Stomach

Stomach

Ink on Washi paper · 2′ × 4′ · 2021
Sketch 1

Sketch 1

Alcohol markers and ink · 2020
Sketch 2

Sketch 2

Alcohol markers and ink · 2020
Sketch 3

Sketch 3

Alcohol markers and ink · 2020
Experiment

Experiment

Hand-made pigments on Washi paper · 2020
Woodprint

Woodprint

Woodblock print with ink · 2020
2018 — 2020
Glowstone

Glowstone

Collage, alcohol markers, pen · 8.5″ × 11″ · 2019
Head in the Clouds

Head in the Clouds

Collage, alcohol markers, pen · 8.5″ × 11″ · 2019
Insomnia

Insomnia

Collage, alcohol markers, pen · 8.5″ × 11″ · 2018
Turtle

Turtle

Collage, alcohol markers, pen · 8.5″ × 11″ · 2018
2015 — 2018
Graphite Work 1

Graphite Work 1

Graphite on paper · 12″ × 18″ · 2017
Graphite Work 2

Graphite Work 2

Graphite on paper · 12″ × 18″ · 2017
Graphite Work 3

Graphite Work 3

Graphite on paper · 12″ × 18″ · 2016
Graphite Work 4

Graphite Work 4

Graphite on paper · 12″ × 18″ · 2016
Watercolor

Watercolor

Watercolor on paper · 12″ × 18″ · 2015

Machine Learning

3D Training Data Synthetic Data Embodied AI

notjustchairs.ai: 3D Training Data for Embodied AI

A venture building large-scale 3D data for both embodied AI and creative work. For robotics, the pipeline produces structured, physically grounded datasets, including articulated objects with movable joints exported as URDF, to train models that perceive and manipulate the physical world. For creative fields, it delivers production-ready 3D assets for games, film, and virtual worlds. More coming soon.

Visit Site →
Gaussian Splatting Video-VAE Feed-Forward 3D

latent2splat: Decoding 3D Gaussian Assets from Frozen Video-VAE Latents

Can a frozen, general-purpose video-VAE latent be decoded directly into an explicit 3D Gaussian Splatting asset, supervised by multi-view render loss alone? On Objaverse-LVIS orbit renders, a strict ray-anchored decoder collapses into translucent fog, while a scaffolded RGBD fusion recovers ~21 dB and ~0.96 alpha-IoU on held-out objects. The latent isn't the wall: the scaffold recovers ~9 dB of lost detail, so the real bottleneck is the decoder's representation and the data, not the frozen video latent.

View Project →
Reinforcement Learning PPO Behavioral Cloning

Dense Reward Generation for RL Fine-Tuning

Behavioral-cloning policies plateau from compounding distribution shift; RL fine-tuning can correct them but is bottlenecked by sparse, binary task-completion rewards. I extract waypoints from the BC encoder's own latent space to produce an unsupervised dense reward, then run PPO on top. The residual pipeline yields +7 points over BC on LIBERO-Spatial and trains 5x faster than sparse PPO while preserving pretrained behavior.

View Project →
LoRA World Models V-JEPA2

Energy Landscape Analysis of LoRA on V-JEPA2-AC

Applied Low-Rank Adaptation to Meta's V-JEPA2-AC video world model for robotic manipulation on DROID-100. Measured energy landscapes before and after fine-tuning across 30 held-out trajectories. LoRA caused representation collapse: flat, non-discriminative landscapes, despite the technique's success on LLMs. A reminder that parameter-efficient methods designed for discrete token prediction don't transfer cleanly to continuous-valued latent spaces.

View Project →