PublicLedger
Audit history — Scaling or refining the current LLM paradigm will not produce human-level intell — PublicLedger