LLM interpretability feels like the low-hanging fruit of the early internet and personal computing— rough enough on the edges that non-technical people don't understand it or use it yet, but *just* functional enough that it can be packaged into a simple b2b product.
The nuclei of many new products are just sitting in research papers (for those with access to niche/ enterprise customers); especially with the renewed excitement around open-weight models.
All without building "just another wrapper".
,
, 
