Research4d ago
Pramana Trains LLMs with Ancient Logic
Apple researchers added irrelevant context to mathematical problems, cutting LLM performance by 65%. The new Pramana method fine‑tunes models on Navya‑Nyaya logic, a 2,500‑year‑old Indian reasoning framework. By enforcing a six‑phase reasoning structure, Pramana reduces hallucinations and improves justification, giving practitioners a more reliable tool for evidence‑based tasks. Unlike chain‑of‑thought prompting, it embeds explicit epistemological steps, easing error tracing.