The Biggest Challenge Facing LLMs

One thing that's common among recent statements by leads of all top AI labs is that they've used all the publicly available data for training frontier models.

imo the next frontier is private, proprietary data.

in the short term, bespoke deals may work (see softbank x openai) but in the long term we're going to need:

1. an economic solution, and 2. a tech solution

crypto is going to be the top contender to solve for both of these.

by tokenizing ip, using TEEs and even through multi-server approaches like MPCs - we will see a lot of new capabilities that will address the challenges of monetization of private data for enterprises who own it with reasonably low leakage and trust guarantees.

however, my bull case for quantum computers is ~5 years. There will be a industry-wide frenzy to switch over to quantum-secure cryptography and verifiable computation techniques. Fully Homomorphic Encryption (FHE) will likely be the top contender here.