Observed arrival · 2026-09-12
Dyno Lab, a Home AI Research Workbench
A local AI research lab that combines Mac MLX inference with an experimental Windows NVIDIA GPU pool for inspecting and intervening in model behavior.
Field notes
Dyno Lab uses local MLX inference on Apple Silicon and can connect to a Windows NVIDIA worker through an experimental GGUF pool runtime. Its research workflow includes activation capture, held-out linear probes, sparse autoencoder fitting, and controlled representation interventions, with settings and results saved for later comparison. The homepage cites a Qwen 0.5B walkthrough and a 72-response experiment, while explicitly stating that interpretability readouts do not certify model safety or alignment.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue