Juraj🏴💛🌘's avatar
Juraj🏴💛🌘
juraj@tamersofentropy.net
npub1m2mv...r8p9
I don’t seek rigid structure — I seek resonance Vibe coding, reality bending, cypherpunk visions. Author of Tamers of Entropy: https://tamersofentropy.net/ I like teaching, get my books and courses here: https://hackyourself.io/shop https://juraj.bednar.io/shop (You'll learn skills no one else is teaching!) Podcasts 🎙️: Option Plus - https://optionplus.io/ Reči o živote, vesmíre a vôbec: https://juraj.bednar.io/reci-o-zivote/ Ako vyhackovať otcovstvo: https://otcovia.com/
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
🎙️ Súkromie: aký je správny level paranoje? Jedni majú vypnutý Javascript a Qubes, druhí mávnu rukou, že "aj tak o mne všetko vedia". Obe skupiny sa podľa mňa mýlia, len každá inak. Podcast je o tom, ako sa na to dá pozerať triezvejšie, ako si pre seba určiť náklady a výnosy. A na konci je bonus o tom, čo robia Google Play services na GrapheneOS. A teaser na knižku Praxeológia súkromia Maxa Hillebranda. image
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
I updated my post on encrypted communication between programs and mobile devices. I now use White Noise primarily and SimpleX secondarily, and I talk to my Hermes agent over White Noise too. For that I built Third Ear, a voice-powered White Noise client for Android (open source, free on Zapstore). It is the best way to use speech to talk to your agent I know of. Amazing how White Noise went from "barely works" to "the best way to communicate with your agent".
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
Aktualizoval som blog o šifrovanej komunikácii medzi programami a mobilom. Primárne dnes používam White Noise, sekundárne SimpleX, a cez White Noise komunikujem aj s Hermes agentom. Na to som vyrobil hlasového klienta Third Ear (Android, open source), s plnou podporou slovenčiny v rozhraní aj rozpoznávaní reči cez OSTT. White Noise prešiel z módu "ledva funguje" do módu "najlepšia možnosť na komunikáciu s agentami".
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
A friend told me an observation: "Switzerland vs. the EU feels a lot like West Germany vs. East Germany around the time the wall came down. It's kind of obvious which model works better." Just not obvious to the honchos in Brussels. I'm still amazed at how out of touch with reality they are. I think many of them genuinely believe they're doing their part in making the world better. They're just not intelligent enough to see the structure of things. Regulating AI when you don't have any AI doesn't do much. It's just a competitive disadvantage. There will be good models in China, in the US, in the UAE. And there will be retarded models in Europe. The funny part is that they do get approved in the end. It just takes three to six months. By the time the stamp lands, the model is two generations old. Nothing got regulated. It just got slower than everyone else. European workers will carry the productivity disadvantage. World not saved. But they don't see it at all, it's a total blind spot for brussels bureaucrats. They still live in a world where the EU has a moat and everyone wants access to EU markets. Except you can have access to EU markets while not complying with their retardation rules in other markets. That's what companies have been doing all along. This isn't the cookie banner situation, where the easiest solution is to troll everyone with the pop-ups. Paid models need to determine whether you're in the EU for VAT purposes anyway. So they just serve you either a retarded model or a normal one. What amazes me most is how fast we forget. Half a year ago the models were noticeably worse, and all of us adapted to the new level within a week. A six-month approval delay isn't a small tax. It's not getting a slightly worse model. It's getting last year's model, permanently, while everyone else moves on.
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
I love how cypherpunk/freedom tech converges. Each of these technologies is very cool on its own, but I always said that the power comes when you connect it together to create sort of network effects or different kind of positive feedback loops. First "it's just money", "it's just a social network", "it's just a parallel communication network". Now I do things like new Nostr client that integrates AI through Routstr paid for by Cashu e-cash built on Lightning. Or a navigation app that uses MeshCore and Nostr for road events, distributed through ZapStore (over Nostr) for all the amazing GrapheneOS users out there. Or interacting with self-sovereign Hermes agent through Whitenoise, with my own Third Ear voice transcription app, which is a native Marmot (Whitenoise) client (a group chat over Nostr), which uses Routstr or Venice (paid for by crypto with funny economic model), using uncensored open models, fallback on local inference, with self-sovereign memory, and the agent itself pays through Lightning, if it wants to order something for me. And will probably read this message on Nostr. We have built a freaking good ecosystem that can do stuff the normie economy can't even imagine. A friend, who is a bit left-leaning lately asked me how my liberatarian world view is doing. He was expecting me to be in total depression, because the libertarian political parties are not winning. I explained, that if nothing changed, it would be amazing. We have liberty and libertarian / cypherpunk world is as mature as never before. Cypherpunk future is now. P.S.: Adding two slop images for completeness :)
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
DeepSeek V4.1 Flash 1017 Uncensored: now live on the Venice API deepseek-v4-1-flash-1017-uncensored Oh-my-God! Access it through routstr.cypherpunk.today @routstr
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
Turns out you can just do things. But when you realize this, you start doing many things.
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
Mal som generátor QR kódov, ktorý sa mi páčil. Nikdy sa neaktualizoval a bežal len ako build pre Macy s Intelom, tak som chcel novú aplikáciu. Postavil som ju dvakrát: najprv len s bezplatnými modelmi, ktoré nepotrebujú kreditnú kartu, potom s DeepSeek 4.1 Flash, ktorý bol na OpenCode Zen zadarmo. Prvý beh trval hodiny, modely sa stále menili, bezplatné kvóty sa míňali a schránka nikdy nefungovala. Druhý beh funguje. Tokenová chudoba, bezplatné tarify a čo vedia a nevedia: Aplikácia: Zapstore:
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
I had a QR code generator app I liked. It was never updated and only ran as an Intel Mac build, so I wanted a new one. I built it twice: first only with free models that need no credit card, then with DeepSeek 4.1 Flash, which was free on OpenCode Zen. The first run took hours, models kept changing, free quotas ran out and the clipboard never worked. The second one just works. Token poverty, free tiers and what they can and cannot do: The app: Zapstore:
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
I wanted to experience if it is possible to build something using only LLMs available for free. I used to build a replacement for an old app that I used. What I used was an old QREncoder app, that is obsolete and built for Intel Macs. So I took a few screenshots, gave opencode v2 a task to build it with a few changes, used impeccable skill to do design (although that did not work very well with the free models). I wanted to know, if anyone can use some sort of machine intelligence, without credit cards, accessible to whoever. Observations: - free models are super slow (auto:smart) - the models change all the time. Forget caching even, it is quite likely, that the whole project is done by 3-5 different models in parallel. I suspect it might not necessarily be a bad idea, since these are not frontier models. - the gap to frontier (even deepseek 4.1 flash 1017) is huge. I almost forgot how much babysitting these models need. Not even /goal helps. - but to my surprise, the app opened and it mostly worked (clipboard and save as png did not work, the icon was generated, but not set). App will be here, but I don't recommend it yet :).
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
Everyone seems to be hacking everything, everywhere, all at once with Kimi K3.1 Pro. image
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
If you have tips how to deal with creativity/productivity/task switching/brain overload happening to us during our use of the AI magic wands, drop tips. I'll start - workout, sauna, walk outside. Preferably interacting with other people, not agents. Thai massage (normal, not happy ends). I wanted to say "Netflix and chill", but I have trouble finding something that is interesting when I can be burning tokens. Any other tips? BTW: Patching remotely exploitable vulns is relaxing, but does not count!
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
What do Google Play services actually do on GrapheneOS, and should you disable them or just restrict them? I wrote this for @GrapheneOS users deciding whether to install them. Typical analyses answer only part of that: they look at the installation package or at the network traffic of one device. But Play services are just a host that downloads further modules based on a configuration from Google's server, so there is no single Play services that could be audited once for everybody. So I also took apart the delivered modules themselves: 24 files pulled from a rooted emulator, and for each one I worked out where it connects, which permissions it inherits and what it could target. As far as I know, nobody has published that. For example, the map and ad modules run inside the app's own process, so denying Play services the network does not stop them. The article also has practical GrapheneOS advice: which permissions to take away, how to reroute location, and what breaks if you switch them off completely. Article: Audio version (MP3): Technical teardown (PDF, 35 pages): https://juraj.bednar.io/wp-content/uploads/2026/10/play-services-grapheneos-GMS-privacy-teardown-v3-1.pdf
Juraj🏴💛🌘's avatar
Juraj 2 weeks ago
Čo vlastne robia Google Play services na GrapheneOS a treba ich zakázať, alebo stačí obmedziť? Na to som hľadal odpoveď pre ľudí s GrapheneOS, ktorí zvažujú, či si ich nainštalovať. Bežné analýzy na to odpovedajú len čiastočne: pozerajú sa na inštalačný balíček alebo na sieťovú prevádzku jedného zariadenia. Play services sú ale len hostiteľ, ktorý si podľa konfigurácie zo servera Googlu stiahne ďalšie moduly, takže neexistuje jedno Play services, ktoré by sa dalo raz zauditovať pre všetkých. Preto som rozobral aj samotné doručené moduly: 24 súborov, ktoré som vytiahol z rootnutého emulátora, a pre každý zistil, kam sa pripája, aké oprávnenia dedí a na čo by sa dal cieliť. Pokiaľ viem, nikto také niečo nezverejnil. Vyšlo napríklad, že moduly pre mapy a reklamy bežia v procese aplikácie, takže zákaz siete pre Play services ich nezastaví. K tomu je tam praktický návod pre GrapheneOS: ktoré oprávnenia odobrať, ako presmerovať polohu a čo sa pokazí, ak ich vypnete úplne. Článok: Audioverzia (MP3): Technický rozbor v PDF (35 strán): https://juraj.bednar.io/wp-content/uploads/2026/10/play-services-grapheneos-GMS-privacy-teardown-v3.sk_-1.pdf
Juraj🏴💛🌘's avatar
Juraj 3 weeks ago
If you ever want to use a microphone of a Bluetooth device on Android, please know that it is dark magic, you have to take chicken paws, under full moon, stir well at no more than 40 degrees and sing kumbaya. Or you can give this to your agent: It is total insanity. Fixed open-source software incoming, because almost no one got it right.
↑