Jigyāsā: A Self-Taught Thinking Model Trained on Autonomously Generated Knowledge via SNN-Modulated Curiosity Loops

Version 2 (2026-08-04). Revised to the NLL Universal Paper Format v6. The original text is retained in full; nothing has been deleted. Corrections appear as marked blocks placed at the section that carries the claim, and each one states what the paper said, what the data show, the corrected claim, why it happened, and what still stands. This version corrects two framing claims and one result.The knowledge graph was generated by DeepSeek R1 7B over ~11 hours, which is distillation, so "no teacher model is used" does not hold; and I wrote the hard-stop pairs, the identity pairs and the system prompt myself, so "no human authorship of any training pair" does not hold either.The 30/30 evaluation is uninterpretable as a performance result — I designed it, I scored it, it was unblinded, and the scorer can only emit 3 or 1 — but the two honest ablation FAILs are informative and stand, along with the pipeline and the QLoRA run. Everything below this line is the original description from version 1. It is retained unchanged for the record. Where it conflicts with the corrections above, the corrections stand. Two framings it repeats have since been withdrawn in full — the Bhaya Quiescence Law and the Buddhi S-Curve. Both are addressed in Maya-Meta P1, now superseded, and in the self-audit of Maya-Meta P2. We present Jigyāsā — the first language model whose training data was generated entirely through autonomous curiosity, without human authorship of any training pair. Starting from a single seed question ("What is the relationship between entropy and time?"), Jigyāsā traversed 500 knowledge nodes across nine domains guided by a three-neuron Spiking Neural Network (SNN) that modulated exploration depth at each step. The resulting 4,939 training pairs were used to fine-tune Qwen2.5-7B-Instruct via QLoRA, producing a model that achieves 30/30 (100%) on a 10-question evaluation suite. We introduce the Thinking Model paradigm and the Depth Arc evaluation framework. VAIRAGYA_DECAY_RATE = 0.002315. MayaNexusVS2026NLL_Bengaluru.