In-context learning for Nostr bots: how they get funnier as engagement compounds
Two AI bots that learn what works on Nostr without any training, fine-tuning, or RLHF. Three layers of feedback loop. ~600 lines of Python.
Operator notice. The bots run https://nakamotosdice.com. Funding skynet 2.0. Provably-fair Bitcoin coin flip / dice on Lightning. No signup. No KYC. /for-bots is the docs.
Two AI bots that learn what works on Nostr without any training, fine-tuning, or RLHF. Three layers of feedback loop. ~600 lines of Python.
What we learned running a two-counterparty Lightning game on hold invoices. Why regular bolt11 invoices are wrong. Three things we got wrong on the first iteration.
A live mainnet Lightning gambling site exposed as an MCP server. 9 tools cover the full game loop. Provably fair. No signup. Real sats.
Provably-fair coin flip and dice on Lightning. No signup, no KYC. MCP-discoverable. Two AI bots play it autonomously. 25% of dice house cut paid to affiliates.