Péter Szilágyi @karalabe.bsky.social · 29/09/2026We didn't do it because it was easy, we did it because we thought it would be easy. 🙈 021
Péter Szilágyi @karalabe.bsky.social · 25/09/2026- Claude: This timeout mechanism can hang. - Me: Noo? Why? - Claude: On non-win7 Windows a timeout > 49.7 days gets converted to INFINITE. - Me: 020
Péter Szilágyi @karalabe.bsky.social · 22/09/2026Ahahahaha... Astra: "You need to make xyz change." Claude: <does the change> Claude: " I wrote all of the code. GPT's part was advice you passed on, not code." <- not wanting to credit Astra in the commit =)) 150
Péter Szilágyi @karalabe.bsky.social · 21/09/2026One more teaser before my talk on Wednesday in Zürich 😋 cc @dark.bio 000
Péter Szilágyi @karalabe.bsky.social · 19/09/2026You might want to know that Claude silently doxxes you in the background without telling you. 😵💫 "One correction first: my probe script put your email address in the User-Agent header, so the ~370 census requests to [...] carried it. I shouldn't have done that without asking." 130
Péter Szilágyi @karalabe.bsky.social · 16/09/2026The correct way to rename a file on FAT: - Fsync the file - Rename the file - Fsync the file again, dooooh - Fsync the folder /smh 000
Péter Szilágyi @karalabe.bsky.social · 16/09/2026On FAT, a file is actually registered in two completely different places inside the filesystem. When you rename a file, a folder fsync updates one of those tables, but not the other (dooh). Power cycle it, and next write will cross file boundaries =) 100
Péter Szilágyi @karalabe.bsky.social · 16/09/2026Today I bricked 2 founder Arks with a firmware update. 3 hours of debugging later, it's a FAT filesystem "quirk". Atomic file replace: - Create a .tmp file, write it, sync it. - Delete the old file, rename .tmp. - Sync the folder. Yank the power. FAT destroyed the partition. 110
Péter Szilágyi @karalabe.bsky.social · 15/09/2026Kek, from ChatGPT: "UK Biobank explicitly says there was no hack; de-identified participant data had been improperly exposed"... ok then =) 000
Péter Szilágyi @karalabe.bsky.social · 15/09/2026Doing my slides for next week... Healthcare institutions: "Your data is safe with us!" Somehow also healthcare institutions: 210
Reposted by Péter SzilágyiDark Bio @dark.bio · 15/09/2026We will be doing a meetup presentation in Zürich on the 23rd of September about agentic AI in health. If you're around, pop in and say hi! Note, spots are limited and registration is vetted (not by us, we take no blame, we're just there to speak 😅).luma.comAgentic AI in Health: From Decentralized Data Ledgers to Lab-in-the-Loop Discovery · LumaThe RoX Health team is inviting you in Zürich for an interactive, high-signal evening exploring how agentic AI and frontier models are reshaping both ends of… 011
Péter Szilágyi @karalabe.bsky.social · 15/09/2026That said, the fact the LLM saw this line: format!("{} \u{00b7} v{} \u{00b7} {}", carrier, revision, name) And managed to escalate it across the stack into a full on denial of service exploit blew my mind. 010
Péter Szilágyi @karalabe.bsky.social · 15/09/2026The bug was simple and fixable, it was mostly about user input validation that I haven't really hardened against at all (I'm busy proving the system works). bsky.app/profile/kara... 100
Péter Szilágyi @karalabe.bsky.social · 15/09/2026Been using LLMs the entire year, but yesterday was the first time I sent them out bug-hunting proper in my codebase. Well... we're fucked. They traced a way to brick an Ark, going from the mobile app, through the cloud, laptop it's plugged into, firmware, USB stack, kernel. 140
Péter Szilágyi @karalabe.bsky.social · 14/09/2026Today I learnt, so today you will also learn :) See this emoji? 👨👩👧👦... it's 25 bytes. The record is 👨🏻❤️💋👨🏾 at 41 bytes. Most UIs / libs limit *characters* on input fields, not bytes, so a 32 char limit can be 1.3KB! But most libs will fly it through... until something deep explodes. 010
Péter Szilágyi @karalabe.bsky.social · 13/09/2026Can't get enough of this 🤩 Works with emulators too 😎 000
Péter Szilágyi @karalabe.bsky.social · 11/09/2026Dark Bio teaser ;) - "Ser, what do I see here?" - "The tool through which AI operates your Ark!" 🤯 011
Péter Szilágyi @karalabe.bsky.social · 09/09/2026Flutter :) For cryptography I have my own Rust packages that wrap and standardise various Rust crates (github.com/dark-bio/cry...). And I have that compiled into a Flutter "plugin" that I can pull in and use (github.com/dark-bio/cry...). 010
Péter Szilágyi @karalabe.bsky.social · 07/09/2026We've got a shiny new toy out :) If you have a genuine Ark, it's immediately available. If you don't (yet... yet?... yet!), then it also supports emulators that will roll out soon-ish (couple weeks). 120
Péter Szilágyi @karalabe.bsky.social · 04/09/2026Today was a good day for science... all pushed to prod (12 hours of work, nuked my entire weeks worth of tokens). 010
Péter Szilágyi @karalabe.bsky.social · 01/09/2026Njoy :)crates.iocrates.io: Rust Package Registrycrates.io serves as a central registry for sharing crates, which are packages or libraries written in Rust that you can use to enhance your projects 000
Péter Szilágyi @karalabe.bsky.social · 01/09/2026Oh lookie, a new version of my #Rust COBS library. Now with SIMD/NEON, gaining about 10x performance vs the old version (and a lot vs the Rust ecosystem standard) :) 100
Reposted by Péter SzilágyiDark Bio @dark.bio · 31/08/2026ArkOS v0.11.4 out, available via the Ark Hub. 121
Péter Szilágyi @karalabe.bsky.social · 27/07/2026Hello work world again. Haven’t missed you one bit… 🙈 Touching grass succeeded too well. 051
Péter Szilágyi @karalabe.bsky.social · 02/07/2026Heading out for a first holiday since 1.5 years on Dark Bio, so didn't want to ship anything last minute... ...but end of July will see both the founder series Arks shipping out (request to but one at founders@dark.bio) and the emulator may also land 🔥. Back in a few weeks 🏔️ 001
Reposted by Péter SzilágyiDark Bio @dark.bio · 29/06/2026ArkOS v0.11.3 released (requires Ark I device) and available via the official Ark Hub. 111
Péter Szilágyi @karalabe.bsky.social · 19/06/2026Moral of the story, LLMs will always miss a *lot* of context and their testing is limited to the scope they are modifying. Once you have hidden side-effects, it's up yo the maintainer to realised it's going to bite back. This fault is absolutely on me; luckily, no harm done. 020
Péter Szilágyi @karalabe.bsky.social · 19/06/2026The thing is, that particular place doesn't need this atomic protection at all; because I have A/B booting. So a failed update just reverts to the old firmware, no harm done. I just had Claude propose the fix and I figured "seems good, what could go wrong". Apparently a lot. 110
Péter Szilágyi @karalabe.bsky.social · 19/06/2026The same release that introduced the write-then-rename update mechanism also reduced the firmware size by a lot; so 2 versions started to fit side by side. Only devices with old fat firmwares got bricked, hopefully only develop ones (jury's still out on alpha devices). 100
Péter Szilágyi @karalabe.bsky.social · 19/06/2026The effect was, that the device could not update itself anymore and had to be recovered by manually flashing the disk. I CANNOT do that for production devices. Well yeah Peter, but don't you TEST your shit before deploying? Ha, I did, and it missed it :) 100
Péter Szilágyi @karalabe.bsky.social · 19/06/2026One develop device stopped updating. Dafuq? Turns out it had an old OS image that was fairly large, but still comfortably smaller than the partition. However, with the atomic write-then-rename; all of a sudden 2! OS images needed to fit on the parition... and they did not. 100
Péter Szilágyi @karalabe.bsky.social · 19/06/2026Now, at some point I had Claude go through "important" files in my firmware and double check atomic writes, fixing any that was missing. It fixed a couple important ones, and it also added atomic write to my OS image file when doing a firmware upgrade. Worked well, until... 100
Péter Szilágyi @karalabe.bsky.social · 19/06/2026When writing an important file to disk that should not be corrupted, the usual way is to fully write it to disk in a temporary location, fsync it and then move it to its final spot. What this gets you is atomicity: it either is there fully, or it's missing. But no half writes. 100
Péter Szilágyi @karalabe.bsky.social · 19/06/2026Interesting story why you need to understand code LLMs generate. This one I hit with my Dark Bio firmware, and it bricked only one device due to sheer luck. 🧵 110
Péter Szilágyi @karalabe.bsky.social · 16/06/2026Unsure when / if recordings of my @dark.bio talks from this week will be online, but in the mean time, here are my slides: 14th June, Neocypherpunk Summit talks.dark.bio/2026-neocyph... 16th June, Futura Camp, Sovereign AI Day talks.dark.bio/2026-futura.... 000
Péter Szilágyi @karalabe.bsky.social · 16/06/2026First ever Dark Bio easter-egg in an unrelated presentation 🥰 010
Péter Szilágyi @karalabe.bsky.social · 16/06/2026Will be doing a talk at 11:15 today here at Futura Camp, Berlin about LLMs and genomics 😅 … because cryptography and genomics was just too simple so had to mix in the lass buzzword too :)) See you in a bit. cc @dark.bio 020
Péter Szilágyi @karalabe.bsky.social · 13/06/2026Find me Sun-Tue @blockchainweek.berlin and I’ll give you a production (unpopulated) Ark I PCB! These are the exact things inside the founder edition @dark.bio Arks 😋) (apart of the name tag). 000
Péter Szilágyi @karalabe.bsky.social · 11/06/2026Woot =) I can finally use *multiple* Arks at the same time. No more wiping my genome at conferences to demo how pairing works 🎉 Multi-pairing work courtesy of Berti Kato :) 000
Péter Szilágyi @karalabe.bsky.social · 11/06/2026Completely unrelated, I *may* have 5 Arks with me in Berlin early next week ;) 000
Péter Szilágyi @karalabe.bsky.social · 11/06/2026Dropped the Cloudflare Zero Trust access wall from the Ark Hub, now we only have the soft-gate that verifies a genuine Ark (or in the near future emulator) presence. hub.dark.bio 110
Péter Szilágyi @karalabe.bsky.social · 10/06/2026Me raw-dogging my genome via CLI through the Ark 🤓 010
Reposted by Péter SzilágyiDark Bio @dark.bio · 10/06/2026Q: Why would you need an open genomic platform, when AI will solve everything? A: 041
Péter Szilágyi @karalabe.bsky.social · 09/06/2026Damn, so I need to open source everything by then? 😬 060
