/muscles
Seven core capabilities. Open each section to see the related builds.
Implementation Depth
Not just 'can you code' but can you take a paper and faithfully reproduce it, debug when it does not work, and explain every design choice.
Feb 19, 2026
World Models 101A ground-up reading of Ha & Schmidhuber's World Models — the MDN-RNN memory architecture, the dream environment trick, and a minimal reimplementation in PyTorch.
Research Taste
Read a paper and identify what is strong, what is weak, what is missing, and what matters. Know which experiment to run next and when a result is real versus noise.
Feb 19, 2026
World Models 101A ground-up reading of Ha & Schmidhuber's World Models — the MDN-RNN memory architecture, the dream environment trick, and a minimal reimplementation in PyTorch.
Mathematical Rigor & Reasoning
Not just knowing formulas, but understanding derivations. Be able to explain why something works, not only that it works.
Feb 19, 2026
World Models 101A ground-up reading of Ha & Schmidhuber's World Models — the MDN-RNN memory architecture, the dream environment trick, and a minimal reimplementation in PyTorch.
Experimental Design
Run the hypothesis -> test -> update loop with the right baselines and clean ablations, and know when you have enough evidence.
No builds yet for this muscle.
Systems Thinking
Understand distributed training, GPU utilization, data pipelines, and serving. Know why a model is slow and how to make it faster.
No builds yet for this muscle.
Communication
State conclusions first, then explain. Surface uncertainty explicitly, and communicate clearly to both technical and non-technical audiences.
No builds yet for this muscle.
Original Thinking
Go beyond reproducing others' ideas: spot gaps in the literature, connect concepts across subfields, and propose your own direction.
No builds yet for this muscle.