Evis Drenova(@evisdrenova):@typesafeai has shown with Jev that most people use LLMs as a universal interface despite that, sometimes, the underlying problem has a narrower structure and that somebody can create a new (and constrained!) model to solve that problem in a much better way than relying on LLMs so then the next obvious question is: what other problems are LLMs solving that can be better served by a new or different model architecture that unlocks more capabilities due to speed/accuracy/cost improvements?

Evis Drenova

@evisdrenova

Principal Eng @ Entire | prev. co-founder @ Neosync (acq. Grow Therapy) @ycombinator @uchicago

Joined May 2019

951 Following 5.5K Followers

Evis Drenova@evisdrenova

2026.09.17 05:04

@typesafeai has shown with Jev that most people use LLMs as a universal interface despite that, sometimes, the underlying problem has a narrower structure and that somebody can create a new (and constrained!) model to solve that problem in a much better way than relying on LLMs so then the next obvious question is: what other problems are LLMs solving that can be better served by a new or different model architecture that unlocks more capabilities due to speed/accuracy/cost improvements?