Latest posts
The ten most recent publications.
For my birthday I was given a book designed by an AI, with my photo on the cover and the title “Die Matrix ruft – Peter antwortet später” (“The Matrix is calling – Peter will answer later”). You might think an AI gift is impersonal, but the personal part lies in the idea, the photo and the punchline, not in setting the letters. This post explains why I see AI as a cordless drill and where the comparison still falls short.
Read post →
Live from Berlin’s Alexanderplatz, Mark Benecke summarises new environmental measurements from autumn 2026: WWF’s Living Planet Report, accelerated global warming, the super El Niño and record lows at Lake Constance and on the Rhine. The core message is that unlimited growth is physically impossible and that everyone can act immediately. Most of the measurements are well founded, but the talk overstates several figures considerably.
Read post →
In October 2026 I installed OpenWrt on my WD My Cloud EX2 Ultra, whose manufacturer support ends in December 2027, and first sent it into a boot loop. A Flipper Zero acting as a serial console saved it, loading the rescue system into RAM via YMODEM. Today it runs as an encrypted RAID1 NAS with its own fan and LED control, and this post contains all scripts and commands to rebuild it.
Read post →
In October 2026 the firewall log of my home network suddenly showed hundreds of connection attempts to my reverse proxy, all from two networks of a Polish provider. The analysis revealed crafted SYN packets whose replies nobody ever answered, so most likely not an attack on me but on the network of the supposed senders. This post shows how to recognise SYN-ACK reflection, why my country block did not apply and how I drop the packets in OPNsense for a limited time.
Read post →
On Hackaday, Maya Posch shows how LineageOS brings Android 15 to a 2016 Xiaomi Mi 5 and turns a Raspberry Pi 4 into a home-built smart TV replacement running Android TV. With the phone, the unlocked bootloader and a missing fastboot driver took most of the time; the actual installation via fastboot and adb sideload then went quickly. On the Raspberry Pi it was enough to write an image to the SD card, and hardware video decoding worked straight away.
Read post →
On 5 October 2026 Rui Ueyama released mold 3.0, the first version of the fast linker after its complete rewrite from C++ to Rust. The new version is meant to be a drop-in replacement for mold 2.42.1, with the same options, target architectures and speed. The goal of the 3.x series is to close the remaining gaps to GNU ld so that Linux distributions can use mold as their default linker.
Read post →
Strata brings Qwen3.8-Flash-Next, a language model with 125 billion parameters, to an ordinary gaming PC with 12 GB of graphics memory and 32 to 64 GB of RAM. Version 0.1.39 was released on 4 October 2026 for the project, which has only been public since 24 September and already had around 11,000 stars on GitHub on 5 October. This article explains what Strata is, how it works technically, what you need it for and where its limits are.
Read post →
On German Unity Day 2026, Aleph Alpha released Kolibri, a language model for German and English with open weights under the Apache 2.0 licence. The mixture-of-experts model has 78 billion parameters, of which around 3.5 billion compute per token, and handles up to one million tokens of context. It is intended for public administration, industry and aerospace, which want to run models on their own servers and be able to trace them.
Read post →
From 2 to 6 October 2026, the International Space Station transmitted twelve images via SSTV on 437.550 MHz as part of ARISS Series 33. I received them in Meerane with a Raspberry Pi, an RTL-SDR stick, a simple magnetic-base antenna and OpenWebRX+, and on the first usable pass two images came through cleanly. The deciding factor was not the antenna, but a small script that tracks the frequency during the pass and saves the images itself.
Read post →
In September I made AI speak locally, got a 35-billion-parameter model running on a single graphics card and, according to OpenAI, used 2.9 billion tokens in the cloud. A present for a friend still did not get finished. This post shows where the waiting time of a voice AI really comes from, why an HTTP 200 can lie and what a token counter reveals about AI agents.
Read post →