Taylor Kolasinski

Founder of Poisson Labs. ML systems & research, reinforcement learning, edge AI. Previously Matter AI, Better.com, WeWork. Brooklyn.

Writing about reinforcement learning, large-scale model training, and simulation.

Notes

View all

May 6, 2026 Where V-JEPA 2.1's Dense Features Hold Up (and Where They Don't) — A pre-registered V-JEPA 2.1 robustness study across all four model sizes, with deployment lessons.
Jan 16, 2026 10,924x: The Instability Bomb at 1.7B Scale — Part 2 of the mHC reproduction series. I scaled from 10M to 1.7B parameters and watched Hyper-Connections hit 10,924x signal amplification. That's not a typo.
Jan 11, 2026 DeepSeek's mHC: When Residual Connections Explode — DeepSeek replaced standard residuals with massive Hyper-Connections, and then watched them explode. I reproduced the 9x signal amplification to understand their fix: the Sinkhorn-Knopp algorithm. Part 1 of 2.

Viz

View all

Visual evidence from reinforcement learning, simulation, and large-scale model training.

Cable Mind: UR5e Demo

The Abyss (2,500 Agents, Unreal)

Amax Explosion @ 1.7B

Cable Mind: 4-Camera Montage

Logs

View all

Apr 2, 2026 Argus: Crack Detection Pipeline
Apr 1, 2026 Argus: Bridge Inspection RL Training
Mar 28, 2026 Stratum: Cognitive Synthesis & Scientific Lockdown
Mar 28, 2026 Stratum: Phase 2 Baseline Implementation
Mar 27, 2026 Stratum: Phase 2 Research Pivot
Mar 25, 2026 Stratum: Bug Fixes & Game Mechanics
Mar 24, 2026 Stratum: Design & Engine Build
Mar 23, 2026 DeepSky ATC: MLP Scaling Limits Identified
Mar 22, 2026 DeepSky ATC: MLP Baseline Refactor
Mar 21, 2026 DeepSky ATC: 50-Agent Coordination Verified
Mar 20, 2026 DeepSky ATC: Stage 4 Production Training
Mar 15, 2026 DeepSky ATC: Reward Function Breakthrough
Feb 12, 2026 Cable Mind: Vision Training & Video Pipeline
Feb 11, 2026 Cable Mind: Camera Placement & Scene Design
Feb 10, 2026 Cable Mind: UR5e Competition Environment
Jan 21, 2026 MARL Cooperative Navigation: Geometry Problem & The Abyss
Jan 18, 2026 MARL Cooperative Navigation: Reward Hacking & Phase 1 Solved
Jan 17, 2026 MARL Cooperative Navigation: Reward Shaping and Local Optima
Jan 16, 2026 MARL Cooperative Navigation: Gridworld to MAPPO
Jan 15, 2026 DeepSeek mHC: The Bomb
Jan 14, 2026 DeepSeek mHC: Infrastructure Hell
Jan 10, 2026 DeepSeek mHC: The Stream Persistence Bug
Jan 9, 2026 Cable Insertion RL: Coordinate Frame Bugs & Reward Ablations
Jan 9, 2026 mHC: Depth Sweep Validation
Jan 8, 2026 Cable Insertion RL: MuJoCo Environment Design
Jan 7, 2026 mHC: Stream Persistence Fix
Jan 6, 2026 mHC: Sinkhorn-Knopp Implementation
Jan 1, 2026 VL-JEPA Reproduction
Dec 15, 2025 Edge AI Platform Architecture