BTZSC

Skepticism Over Jev: Old Tech Marketed as New AI Paradigm

A critique argues Jev is not a new class of decision model: its zero-shot classification behavior already exists in NLI classifiers, embedding models and rerankers, and it cannot generate autoregressively or emit invalid classes. Marketing comparisons against LLMs tout speed and cost but lack benchmarking on suites like BTZSC; in one Banking77 test, BGE-small plus logistic regression scored 93.3% versus Jev's 83.2%. The '0% hallucination' claim only guarantees outputs match the allowed schema, not correctness.