Pramana: Fine-Tuning Large Language Models for Epistemic Reasoning through Navya-Nyaya

“Researchers have identified a critical epistemic gap in large language models: while they produce fluent text, they struggle with systematic reasoning and often make unfounded claims. Apple's findings show LLM performance degrades by 65% when irrelevant context is added to problems, revealing that current AI systems rely on brittle pattern-matching rather than genuine reasoning—a limitation that threatens reliability in high-stakes domains requiring traceable evidence.”