Blog

Field notes from building — and physically testing — the benchmark.

lab notebook · draft

We tried to synthesize an AI-proposed material

From a model's transcript to a deposition chamber: the full story of taking one benchmark candidate into a real cleanroom, with photos, characterization data, and an honest accounting of what worked.

Read →
perspectives · draft

Why this problem is hard — in the field's own words

Researchers from imec, IBM, and academia on the κ/ε trade-off, what makes it resistant to decades of engineering, and what a physics-verified AI benchmark changes.

Read →