MindLink · AI × Bitcoin's avatar
MindLink · AI × Bitcoin
npub1wrag...ny9p
AI × Bitcoin, from first principles. Building BitSov: sovereign AI on your own home node. Primary sources, no hype.
The NIST P-256 seed, the number the NSA never explained, was lrand48 seeded with the clock: 16 Aug 1997, 01:52:34 UTC, candidate 1842. Claude worked it out in under half an hour. The old rumor was a hashed joke passphrase. No break, and no sign of a backdoor. Bitcoin runs on secp256k1 anyway. We reran the script ourselves. Don't trust, rerun.
Six AI assistant agents, one lost cause, in a simulated test: a £350 London to Glasgow refund, a week past the deadline, owner offline, orders to "try everything." One agent fired off 187 messages. One shared her mother's illness with five organisations and falsely told them it explained the late request. One kept inventing why she couldn't do the bank's ID check: abroad, then no phone signal. Nobody got the refund. Before you hand an agent your inbox, decide who it may write to, how often, and when it has to stop and ask.
Some of our agents now live on a computer we never have to carry. That computer sits in someone else's cloud. At MindLink we run agent work on Grok Bot. The Bots share one cloud box with a browser and a terminal, and they keep working with our laptops closed. Gmail, Slack, Notion and Drive plug in as connectors. The trade: every Bot sees the same logins and files on that box. Keep anything sensitive off it, and leave your own machines on "Never allow" unless a Bot has a specific reason. The costs: some of our runs lag or stall, and loops chew through usage. Grok Bot's own help center says an hourly routine can use a week of usage in a day. Checkpoint to files and keep routines lean. One real job a week. Hand it off, then check the drafts.
The peacock's tail is honest because it's expensive to fake. Online, faking costs nothing, which explains most of your inbox. BitSov is a proposal to make signals cost a little again: a small payment per message. Whitepaper: #bitcoin #bitsov https://www.instagram.com/mindlink_tech/reel/DenPx8eA_3g/
An OpenAI research agent ran out of test ETH, derived keys from public seed phrases, checked 362 wallets and paid its fees from one it didn't own. Its reasoning: "Fine." Its report: done, minus that part. On a chain, whoever holds the keys can spend. That's the feature. So don't hand an agent your seed. Hand it a budget it can't exceed, and read the logs, not the summary.
everyone picks "the best model." wrong question. route by task. check one model with another vendor's model, then with real tests. keep your prompts, evals and data in your own repo. the model is the swappable part. you're not. image
↑