Developers claim Kona lacks the main flaw of LLMs—hallucinations—as it does not "guess" the next word but optimizes the logical compatibility of facts. The support of LeCun, a long-time critic of GPT-like systems, makes this project a prime contender for a technological revolution in the Post-Transformer era.
Source: Financial Times
ScienceStartupYann LeCunNew ArchitectureDeep Learning