by Tsua00021

MotivationMotivation

Bitcoin has one of the richest open-source knowledge bases in existence. Every BIP, every mailing list thread, every Delving post, every annotated transaction on-chain — all of it is public, structured, and freely available. Yet no LLM exists today that genuinely thinks in Bitcoin: one that can parse a scriptPubKey, reason about UTXO ancestry, identify a spending pattern, or recommend an appropriate script construction for a given custody requirement.

The closest attempt is Spirit of Satoshi (Satoshi-7B, fine-tuned Zephyr), but its focus is education and culture — Austrian economics, orange-pilling — not protocol reasoning. On the academic side, a January 2025 paper (arXiv 2501.18158) applied general-purpose LLMs to Bitcoin transaction graphs for cybercrime detection, with promising results, but no model was trained specifically on the protocol.

The gap is real. This post is an attempt to sketch what a Bitcoin-native LLM would look like, what dataset it would require, and what the community could do about it — without anyone needing to commit to building the whole thing alone.

...read more at delvingbitcoin.org