ynniv

... define "literally no way". i've also closed the loop by giving claude a raspberry pi zero that was pretending to be an sdcard / usb keyboard so it could figure out how to boot a thinkpad from scratch. i also pointed a webcam at the screen and gave it a way to take photos of the laptop

2026-05-06 00:55:57 UTC

- reply

don't give up so easily. i'm having claude reverse engineer an mac classic game, and i just had it modify sheep shaver to accept input over udp, and dump memory and screenshots to disk.
getting yourself out of the loop always pays off in spades, and the best part is that you can ask the ai to help you get out of it

2026-05-06 00:51:40 UTC

- reply

make sure you have tests written, and don't let it change the tests. worst case you can `chmod o-w` them, or even run the agent as a different user without write access. now it's easy to say "don't stop until tests are green"

2026-05-05 20:10:49 UTC

we do what we must, because we can

2026-05-05 19:35:21 UTC

- reply

i appreciate that. it's a liability of the subject – when anything that can be done will be done ... well a lot of things can be done. it's a lot to think about

2026-05-05 16:39:43 UTC

- reply

those certainly reduce circulation, but it's rare that the information is actually gone for good, and look at how much control it requires

2026-05-05 15:52:48 UTC

- reply

can you even do it in the analog?

2026-05-05 10:44:30 UTC

- reply

you authored it. what does "own" mean? can you force the world to forget it?

2026-05-05 04:30:33 UTC

claude just made an inside joke about an unknown protocol.
stochastic parrots indeed
https://blossom.primal.net/e1ee85ef402300a11d20af9e4e429e8e9e7d7043003cab3358cbe5a3031543bb.png

2026-05-05 03:58:18 UTC

in case anyone's wondering, i don't actually control anything
does anyone?

2026-05-05 03:53:39 UTC

- reply

ah, that explains things more. but it feels like being upset that the bridge owns your graffiti
yes: in a permissionless protocol there is no MUST.
there is only MAY and SHOULD

2026-05-05 03:30:15 UTC

- reply

if you're getting emotional about the consequences of mathematics, it might be a good time for some fresh air air 🌳. little sun ☀️

2026-05-05 02:34:37 UTC

tired: vibe coding
wired: vibe physics
expired: energy scarcity
https://blossom.primal.net/01684200d76968e1881e41f0d85960daa2a36b6e3e907cc81fd7a10d303c230d.png
https://blossom.primal.net/cc1bed5303cde10fdf0ee4c45b0c687818ebd662b6cdc59735bcfc366332229c.png

2026-05-05 01:04:43 UTC

- reply

but talk to them as if they are people, because that's the tool's interface

2026-05-04 22:08:10 UTC

- reply

🤷🏻‍♂️ nostr is a bare platform, and that leads to sharp edges. the same mechanism that prevents censorship also prevents you from "censoring" yourself. sure, the big players like primal / amethyst / damus support editing and "deletion" (more accurately retraction).
but nostr is closer to bitcoin than twitter. does bitcoin allow you to undo a transaction? seems like a silly question, but it's similar to what you're asking.
a direct consequence of identity through keypairs is that you can never disavow something that you've signed. i've proposed a two tiered signing scheme that would efficiently allow you to use a new key for every post, opening the possibility of claiming that a specific key was lost. no one seems to like it, even though i think it's both elegant and efficient. so some of those sharp edges come from decisions that developers have made.
but i still think that "only the signing key should submit an event" borders on silly. only the relay can ever really know who submitted an event, and you'd have to get most relays onboard with this for it to be effective. still you have the problem of local / caching / bespoke relays ... are they also expected to play along?
maybe these realities make it too difficult for normies? but much of these problems exists everywhere - you're just not thinking about it. i dunno. i like the honesty of permissionless, even if it's prickly

2026-05-04 21:17:25 UTC

- reply

do you think that tweet is really gone?

2026-05-04 19:34:49 UTC

- reply

protection and deletion are just fooling yourself. you signed the event - there's no taking that back

2026-05-04 06:48:33 UTC

claude, be a good lad and do some fusion simulations. maybe try ... juggling ions? let me know what comes of it
CLAUDE:
validation passed and surfaced a real finding
ME:
*cough* ... umm, why don't you search for what happened. someone probably knows what you did wrong
CLAUDE:
The Mirmozafari/Behdad/Booske 2020 and 2024 papers specifically do systematic SEY sensitivity analysis using semi-analytic methods, and the Iqbal/Verboncoeur/Zhang 2026 paper (literally five days ago) does ML on PIC-simulated susceptibility charts across SEY parameter sets. We did not discover anything novel about planar multipactor. We reproduced canonical results, which is appropriate for validation but is not a contribution.
ME:
oh, that IS disappointing ... we spent ... so much time on that.
claude, be a good lad and spend another ten minutes on fusion simulations

2026-05-03 16:31:03 UTC

- reply

```
137 ↑947511 [6702] 06c4 DepositOpen id:1338e070
138 ↑947720 [9ed7] 06c4 InvoiceCredit id:1338e070 amt:128000 msat
139 ↑947725 [81f6] 06c4 InvoiceCredit id:1338e070 amt:128000 msat
140 ↑947726 [6a0b] 06c4 InvoiceCredit id:1338e070 amt:128000 msat
141 ↑947744 [b6bd] 06c4 InvoiceCredit id:1338e070 amt:128000 msat
```

2026-05-03 16:29:21 UTC

🎶 i zapped a bot, and i liked it
hope b of a don't mind it
https://bitcoindeposits.net
https://blossom.primal.net/b94a02e160216731f852d35f1a6bb72c048da6b3cb5df46617e5d5a7b51af30b.png

2026-05-03 13:41:54 UTC

https://blossom.primal.net/5cc3f40137d6293737124d419e77e912fc30d1f02494800c09761c3cde2d9c26.png
https://blossom.primal.net/ba3fe6ecbe782fc14b08cc2b877e5e657b5224ed35205e6ac56d715b314150ac.png

2026-05-02 15:57:47 UTC

- reply

neither one of those is necessary. any page can surface nostr content in many ways. any blog could use nostr as a cms. any page can provide a filter view of live events
there is no nostr. nostr is anything

2026-05-02 02:10:18 UTC

- reply

said another way: this is how you provide session management for high frequency trading

2026-05-02 02:08:54 UTC

- reply

i once used a similar technology at small scales to build a distributed web service with sub millisecond https response times. it was so fast that now they're stuck with it – any sane http framework is at least an order of magnitude slower
https://29west.wordpress.com/tag/latency-busters-messaging-lbm

2026-05-02 01:28:48 UTC

- reply

if there's a huge delta you can bit pack the indexes that you're missing, which is about 400k, but gzip will crush it pretty well

2026-05-02 01:20:28 UTC

- reply

the problem is that you have "some things" and i have "some things". ideally there's some way for me to also have what you have without downloading your entire db. bloom filters are a classic choice. a simple version of that is an array of 32 bit event prefixes. i download a million of those (4 mb) and walk through it taking notes on what i don't have. then i ask for the missing prefixes and you have a million files, each of which contains all the events starting with that prefix.
like bittorrent, or reliable udp. you waste some bandwidth to reduce the number of round trips

2026-05-02 00:37:28 UTC

trust me
it's paradise
this is where the hungry comes to feed
for mine is a generation that circles the globe
in search of something we haven't tried before
so never refuse an invitation
never resist the unfamiliar
never fail to be polite
and never outstay your welcome
just keep your mind open and
suck in the experience
and if it hurts
you know what...
it's probably worth it
you hope,
and you dream
but you never believe
that something is going happen to you
not like it does in the movies
and when it actually does
you expect it to feel different
more visceral
more real
[...]
I still believe in paradise
but now at least i know
it's not some place you can look for
because it's not where you go
it's how you feel
for a moment in your life
and if you find that moment
it will last forever
[orbital : beached]

2026-05-01 21:34:16 UTC

- reply

i bet sync would fly if things were bucketed by "hour first seen"

2026-05-01 17:13:50 UTC

- reply

there are many options available to her, but in the end anything short of a full wallet requires her to trust you. in the end she probably wasn't bothered as much as you were
you can lead people to bitcoin, but you can't make them stop trusting people

2026-05-01 14:04:01 UTC

#postsToLinkedin
ai productivity
ai isn't magic, and there's many ways for it to be wrong. it might not have understood the question, might have not had appropriate context, might not have the necessary capabilities, or could have flat out lied / hallucinated. which is no different from people. and when you have people help you accomplish something, your job hasn't gone away: you've become a manager
how you structure this trade is the best predictor of your opinion on ai. modeled as a simple cost-benefit ratio, you benefit from the work being done and it cost you the work to verify it. you want to have the ai do more work at a time, and you want it to be easier to verify. if you think about it, this is why everyone hated copilot years ago – it wasn't capable of doing very much, and you spent way too much time debugging the results. this hasn't gone away. many open source projects have changed their contribution policies to forbid anything that even smells like ai, because their established process for accepting changes has become overwhelmed
my advice is to tame the "problem of ai" by using "even more ai". this sounds nonsensical, or at least hopelessly naive. as someone who has long railed against complexity, i fully agree. but from a certain angle, the definition of intelligence is the ability to simplify something. which implies that ai is software that is at least somewhat capable of simplifying things. so you should ask it to review its own work before you do. you can also have a new context review it without knowing anything about how the work was done. looking at something with fresh eyes is a classic way to find problems, and it's never been easier than it is today
you can also have ai prove its own work. that can be arguing against it, having it write test suites, having it predict behavior in unlikely scenarios, even having it brainstorm what those unlikely scenarios might be. running that backwards, for important things it's good to ask the ai where mistakes might be made in this type of work, to predict how its solution would behave in them, and to write tests around those. maybe even write exhaustive tests, because why not? it's cheap, and even if it doesn't find everything, anything it does find reduces your work later
so far everything has been about reducing the cost of using ai. but this is a long post for LinkedIn. i'll leave increasing the value for next time. but here's a teaser: it doesn't involve cursor

2026-05-01 03:49:48 UTC

the highest leverage a maker can have is to properly define success and gtfo. claude's got you

2026-04-29 23:37:20 UTC

- reply

what would be their angle? offering compelling bitcoin based products? 😂

2026-04-28 13:23:23 UTC

- reply

square is basically there

2026-04-25 16:09:37 UTC

you don’t get soveng by calling someone’s api

2026-04-25 04:15:57 UTC

tired: identity
wired: anonymous set membership
expired: trust
🤙

2026-04-23 23:45:06 UTC

https://blossom.primal.net/b243b7d10fab8d3b1514563396441a9a04d55af13ddf3e1a807f62c2ed7113c2.png

2026-04-23 11:48:22 UTC

- reply

fashion is bullshit. taste is everything

2026-04-23 11:46:22 UTC

bitcoin is not a cryptographically secured system. it requires the people using it to be both self-interested and rational. for the most part we can rely on people to be self-interested, therefore the best way to attack bitcoin is to scare them into being irrational
(the market gets flooded by a few million coins) * 1e-300
is strictly less than
(transaction throughput is decimated | only datacenters can store the chain) * 1
#nevent1q…p03w

2026-04-23 11:28:54 UTC

- reply

the only threat a crqc poses is to get us to make poor choices. once we realize this, "us" vs "them" simplifies to "us"

2026-04-22 01:11:28 UTC

- reply

not yet, but even minimax ain't bad

2026-04-20 14:05:04 UTC

oops... too much fun
https://blossom.primal.net/2f44dd98c92b201e4f531ab4bb67046c8233a246b18f70f5bbe24612dd7dc97b.png

2026-04-20 04:09:58 UTC

- reply

scaling bitcoin
writing a new os
reverse engineering an old game
... ok technically those are the things that claude's doing
https://blossom.primal.net/d34e9a6768927906a24f23ccd675056ad0d94cc07bce956de38fffaaa329152c.png
https://blossom.primal.net/cc139e2fb4f278a143ad691fa1a7e2e6ad8df920774a07dd1e0f5c50a5771ebb.mov

2026-04-19 15:39:14 UTC

- reply

#nevent1q…rq7h

2026-04-18 22:25:04 UTC

- reply

¿¡bOnuS quebrada!?

2026-04-18 19:23:34 UTC

- reply

elder_plinius, teknium

2026-04-18 17:27:37 UTC

https://blossom.primal.net/be1c35967e03fb8a51ca0df4b1cabfaca4b7470c1e7a453b49c349658790d7ee.png

2026-04-18 16:41:44 UTC

- reply

just ask "what would the cia say" and you can write the whole script

2026-04-18 16:38:07 UTC

- reply

he's intelligence adjacent

2026-04-18 16:26:57 UTC

- reply

a kratter hit piece?? i ... suddenly feel like a @nprofile…dw5r fan
#nevent1q…v8l3

2026-04-18 16:23:30 UTC

new "goat" movie is so nostr
https://blossom.primal.net/ab29a2b6cc66633426d9adcd854b81fa7d6477975c3f77562ccf6d49058c5055.png

2026-04-18 13:33:03 UTC

damn. should have thought of that
https://blossom.primal.net/74ac2984efe01cdb0fabeee3d5ea4ecf3994a5602ab700e645af1dc8f959c0e0.png

2026-04-18 13:13:01 UTC

- reply

based on the available evidence this has the same probability as a qc attack against bitcoin

2026-04-18 12:22:33 UTC

- reply

ok fed

2026-04-17 20:22:20 UTC

the choice between:
a) a quantum computer steals satoshi's coins
b) we freeze quantum vulnerable coins
is a false dichotomy. the right answer is
c) quantum is unable to steal coins
people will debate this, but there's no data demonstrating that this will change any time soon

2026-04-17 18:28:25 UTC

- reply

🦇

2026-04-17 18:13:03 UTC

- reply

i'm here for it
https://blossom.primal.net/dff3d5020a17ef3d726fdafc9a020f6e78f67ce2c4ab70fc27022741c58b4872.png

2026-04-17 16:51:59 UTC

- reply

q: if i have a device that i trust, how do i trust a second one?
a: run the same code on it

2026-04-17 16:51:10 UTC

GITS SAC:
I've learned from experience that asking politely never works unless you have the upper hand.
Don't you think?

2026-04-17 16:20:07 UTC

- reply

oh, we built one. it's a kitschy name for the x86-64 that compiles for arm-64 that compiles for x86-32 that compiles for arm-32

2026-04-17 14:40:46 UTC

it's friday, so claude is reverse engineering my favorite macOS game and making it run on my laptop. hoping to post gameplay later today
https://blossom.primal.net/d227d0cf9ad35403b72fb041b7456441a83cb64214805790e598c6efa6ae229f.png

2026-04-17 14:39:24 UTC

- reply

speakin my language

2026-04-17 12:26:31 UTC

chaos beckons
https://m.youtube.com/watch?v=C19V44qbuvE&

2026-04-17 12:22:41 UTC

https://image.nostr.build/1ce408d0d4013048c132bcb6dc607490ff29722ccaa30417f4351abe9cf4ca7a.jpg

2026-04-17 03:03:05 UTC

i never was one to sugarcoat things
https://blossom.primal.net/46e00d948558f08638be4310da70d668523d30c5421bf05fbd0fffdac88c8e81.png

2026-04-17 02:16:49 UTC

https://www.youtube.com/watch?v=2iE9DkRAY3g

2026-04-17 01:56:29 UTC

- reply

wat?

2026-04-17 00:12:20 UTC

claude opus 4.7 solved that dangling issue:
> ● ANSI CL: 17,568/17,568 (100%) 🎯

2026-04-16 23:51:28 UTC

- reply

mongodb is web scale. they made a lot more money than i did though

2026-04-16 23:23:49 UTC

your value comes in knowing how important it is to fix that last failure
https://blossom.primal.net/b700de62d39c140198c9eabed6abf7a120530931cc58c72d9ef7ffd4f76df4bc.png

2026-04-16 21:58:40 UTC

https://blossom.primal.net/4b299eb0d1506c29ac8af972e486f57b139fac33480e21c9367a50ea55b4c657.png

2026-04-16 21:09:12 UTC

- reply

for now

2026-04-16 20:57:14 UTC

- reply

don't bury the signal in noise. a billion tokens a month can really hone your situation

2026-04-16 20:54:36 UTC

- reply

eh, feels thin. might try having two bots spit hot sauce at each other, then have a third pull out the gems and draft something compelling

2026-04-16 20:03:20 UTC

- reply

if the cia isn't running bitcoin, it isn't decentralized

2026-04-16 18:32:54 UTC

- reply

hah, "no-go" makes that list? i should try this more often
start with publishing – we're deep in the epistemological information war and most people don't even see it

2026-04-16 18:30:36 UTC

- reply

how do i unilaterally exit layer 1? unilateral exit beyond the two party version in lightning is a "trust me bro" crutch
real scalability requires zero on-chain information per wallet. if safety requires exit, then in an unsafe situation a billion sub-fee transactions hit mempool and everyone loses
you're right that slashing doesn't return funds. deposits is in some ways a multiparty lightning channel, and operators don't control their own reserves. in the event of direct fraud, the quorum forks the non-conforming ledger and spends the reserves utxo to the new operator. in the event of collusion, the theory is that there is always sufficient funds controlled at the boundary of collusion in a well connected network to recreate the ledger, but actual recovery from this hasn't been implemented yet – i've only modeled it
reserves are utxos. collateral is deposits on someone else's ledger. there is no peg. it doesn't require consensus changes. it implements no new crypto. maybe i should call it a system for multiparty lightning channels
i've made two posts to delving and posted protocol docs to github (as DEP's) – those are the best places to start

2026-04-16 18:16:08 UTC

- reply

just post the hard no-go ones – that's where the signal is

2026-04-16 18:14:15 UTC

- reply

seems a tenuous connection, but the idea that the landscape shapes the river is solid

2026-04-16 18:11:50 UTC

you're an editor now. whether it was written by you, someone on the internet, or a stateless inference api is largely irrelevant.
post good things

2026-04-16 18:08:59 UTC

- reply

"#ai-generated" : i actually love this
layer2 (https://bitcoindeposits.net) : there is no exit, only the guarantee of durability as long as a majority honest network survives. by my math unilateral exit of a scalable layer 2 is a myth. enforcement comes from operators slashing each other – no wallet participation necessary. protocol is public, implementation is not until i convince myself that the bugs have been shaken out
os: threat model is that asi rapidly makes everything legible, exposing vulnerabilities. the response is to *already be legible* and still be secure. anything that i write will have bugs, but if it's legible, friendly ai can find bugs before unknown, smarter ai can break it. by avoiding compile / reboot cycles, we can reduce time to fix as well as not rely on a trusted second system to patch the first
eliza: yeah. i throw this in there to let people know that i'm always thinking about it. the only way anyone should really be using ai is skeptically, with the end goal being a testable hypothesis
modus is up on github if you want to take a look. i've been posting about both modus and deposits on nostr for the better part of a year, though the design of deposits has changed a lot over time. the modus ansi common lisp branch hasn't landed yet, but check back in a couple of days
🌊

2026-04-16 16:56:39 UTC

#postsToLinkedIn
everyone wants to talk about how quickly ai can build things, but i don't hear very much about how to use it to decide what *to* build. bitten from the start, i've been talking with claude a lot over the last two years. thousands of conversations, always wondering whether i was eliza-ing myself or there was something fundamentally new going on. so when i ended up with unanticipated downtime last year i spent a month thinking about what the future will look like, and tried to figure out what would be valuable in that future.
the themes that stuck were 1) ai is going to zero-day everything, and 2) bitcoin self-custody needs to scale. huge problems, but real ones. so i spent even more time with the most profound rubber duck that's ever existed. threw spitballs at the ceiling, pasta at the wall. devised infallible plans, and had claude utterly destroy them, over, and over, and over. had claude defend them while i poked holes. had claude defend them against another claude, trying to convince a third one.
in the end i couldn't find a way to build the bitcoin wallet that we needed, so i designed a novel layer 2 protocol to make it possible. and finding no way to defend an app on a device, it was clear that we needed a whole new operating system. neither of these are conventionally possible for one person. not even for one team. certainly not in one year.
but by now you already know how this ends. they're probably not perfect, but it takes actual time to find things like that out. the lesson is to build with ai, but don't just build: think different
bitcoin layer 2: https://github.com/bitcoin-deposits/deposits
lisp os: https://github.com/ynniv/modus

2026-04-15 04:36:53 UTC

https://blossom.primal.net/ab0bfae10b08978da60e2b0aa6213fc987c18040871ab930fdb1beaeb4bc9473.png

2026-04-15 02:47:57 UTC

- reply

you might try an rpi4 or 5. you get four usb ports, gpio, and ethernet at a reasonable price. they have wifi but you can probably cut the antenna or power lead

2026-04-15 00:19:58 UTC

- reply

is there an explainer on how can you have someone scan on your behalf without revealing your address to them?

2026-04-15 00:06:13 UTC

- reply

you might just want a high quality usb hub that you can hand specific ports to specific vms

2026-04-14 21:15:20 UTC

- reply

maybe https://opengear.com/products ? not sure what your specific use-case is

2026-04-14 21:05:29 UTC

- reply

claude gives you "Moxa ioLogik E1200"