Research4d ago
Pramana Trains LLMs With Ancient Logic
Apple researchers added irrelevant context to math problems, cutting LLM accuracy by 65%. Pramana trains models with Navya‑Nyaya, a 2,500‑year‑old logic that forces six‑phase, evidence‑based reasoning. The fine‑tuned LLMs produce claims that trace back to explicit premises, reducing hallucinations. Practitioners can now deploy models that justify answers, improving trust in high‑stakes domains.