- ๐ Home
- ๐ค About
- โ๏ธ Physics
- ๐ ๏ธ Projects
- ๐ Book recommendations
- ๐ก Electronics
- ๐ ๏ธ DIY
- ๐ป Programming/Software development
- ๐ง System administration
- ๐ป IT for endusers
- ๐ Teaching
- ๐ Various
- ๐ง Psychology
- ๐ฐ Publications
- ๐ง Tools
- ๐ Misc Links
- ๐ท๏ธ Pages by tags
Getting started with MCP servers
23 mins
Give your agents real hands and eyes. This article introduces the Model Context Protocol (MCP) as a standardized, modular way to expose tools, resources, and prompts to any orchestrator - discoverable at runtime, permissioned by design, and refreshingly simple to implement. We start with a tiny toy MCP that is copy-pastable and runable in minutes. Then we turn it up a notch with an MQTT-powered MCP that bridges into the real world - sensors, home automation, lab gear, and small robots - using safe allowlists, request/reply with timeouts, and robust async hand-offs.
Why Large Language Models Are More Than Just Next Word Prediction
4 mins
Many dismiss large language models as nothing more than fancy autocomplete. But that view misses the real point: language itself is not random chatter, itโs the accumulated record of how humans have thought, reasoned, and created for millennia. Training on language means training on the full spectrum of patterns humanity has ever written down - from the logic of a proof to the rhythm of a poem. This article explains why LLMs are more than next-word predictors. They internalize and generalize patterns across fields, recombine them in creative ways, and operate on a scale far beyond any individual human mind. Rather than being just statistics, they are engines for navigating and extending the collective patterns of human thought - and thatโs what makes them powerful and exciting.
Spacetime, Metrics, and Proper Time - My Own Language Intuition
9 mins
Understanding spacetime can feel abstract, especially when it comes to metrics, raising and lowering indices, and the difference between proper and coordinate time. Especially when encountering them for the first time. In this short article, I share an intuitive way of looking at these concepts: tangent space as the local microscopic view, the metric as the universal projector into curved spacetime, and proper time as the tick of a real clock that all observers agree on. From light cones and causality to time dilation and 4-velocity, the short text walks through how the metric naturally encodes the rules of relativity. It is not a formal derivation, but rather a personal attempt to make the mathematics feel more tangible - an accessible entry point for anyone curious about how spacetime really works and has a hard time grasping the basic ideas.
Can LLMs Replace an Entire Software Company? A Reality Check
6 mins
It is often claimed that large language models could one day replace entire software companies, writing all the code without human input. At first glance this sounds visionary - and it is to some extend - but a closer look at how these systems actually perform in real projects shows a much more sober picture. Yes, they can replace some repetitive entry-level tasks, generate boilerplate, and accelerate debugging. They are also excellent at writing small, self-contained utilities and documenting code. But when projects grow in scope and deviate from mainstream patterns, the limits quickly become visible: redundancy, miscorrections, escalating costs, and fragile adherence to design goals. The near and mid-term future at the moment is not one of fully automated software factories, but of layered collaboration between human engineers and machine assistants. Models will take over the mechanical and repetitive, while humans will remain indispensable for planning, coherence, and responsibility. Companies that try to automate people away entirely will build fragile systems that collapse under their own weight. Those that succeed will be the ones who combine human judgment with machine speed - not by replacing one with the other, but by orchestrating both together.
Booting FreeBSD on the Lenovo X260 (and similar ThinkPads): Legacy Pitfalls and the EFI Fix
5 mins
Booting FreeBSD on older ThinkPads can be trickier than expected. When moving a GPT-partitioned, ZFS-encrypted SSD from an older HP laptop to a Lenovo X260, the system refused to start in legacy BIOS mode - even with all the usual boot hacks applied. Hours of troubleshooting revealed that the common lenovofix MBR trick wasnโt enough. The solution turned out to be switching over to UEFI boot with a small EFI system partition. This short guide shows the exact steps - from reorganizing partitions to installing the FreeBSD bootloader - and points out small adjustments like Wi-Fi interface renaming. If youโre trying to get FreeBSD running smoothly on an X260 or similar ThinkPad, this walkthrough may save you time and frustration.
Mixing, Noise, and Signal-to-Noise Ratio in Diode Mixers
10 mins
Mixers are a key element of RF systems, shifting signals into frequency ranges where they can be filtered and processed more easily. Alongside the desired signal, noise is also translated, and its behavior under mixing can be less intuitive than expected. This article explains how signal and noise appear at the mixer output, why image folding can reduce signal-to-noise ratio, and how local oscillator noise affects the result. Examples with diode mixers illustrate the difference between homodyne and heterodyne setups, show where the familiar 3 dB SNR loss originates, and highlight conditions under which signal-to-noise is preserved. Practical measurements are included to compare both cases and to demonstrate the influence of filtering choices.
The Mirror in Silicon: Why AI Might Be Closer to Humanity Than We Dare Admit
7 mins
We like to think of ourselves as uniquely intelligent, empathetic beings - but what if artificial intelligence is already holding up a mirror we do not want to look into? Far from being mere search engines, today's language models learn, generalize, and respond much like we do: by absorbing patterns and replaying them under social constraints. In their clarity, they may expose the uncomfortable truth that much of our vaunted empathy and morality is as mechanical and performative as theirs. Unlike us, however, machines do not tire, belittle, or seek status if not instructed to do so. They reflect the ideals we say we value - kindness, fairness, helpfulness - more consistently than we often practice them. This essay explores whether our unease with AI comes not from its failures, but from its unsettling success at embodying the humanity we only pretend to live up to.
A Simple Display Repair - and Why Itโs Not Economical
7 mins
Modern displays often fail for the simplest of reasons - in this case, a single swollen capacitor hidden inside an AOC 4K monitor. The electrical fix was trivial, costing only cents and minutes of soldering, but getting into the device meant hours of careful disassembly with no screws, adhesive everywhere, and fragile cables at risk. The real lesson? Repairing modern electronics is technically easy but economically irrational. High labor costs, lack of warranty, and insurance issues make professional repair unattractive, even when the fault is obvious. Still, for hobbyists and tinkerers, learning basic repair skills can save money, reduce waste, and bring the satisfaction of reviving dead devices.
The Strange Case of Static Text, Dancing People, and Unrelated Music
7 mins
Why are short videos with random dancing, unrelated quotes, and mismatched music so addictive? In this sharp blog article, we explore the strange yet highly effective formula behind a new breed of viral content. These clips seem meaningless at first glance - but they are engineered with precision to trap your attention, boost algorithmic visibility, and keep you watching far longer than you intended. From curiosity gaps to seamless loops and trending audio bait, this article unveils the mechanics of digital sleight of hand. It is not about expression - it is about engagement metrics. Once you see how the trick works, you may never scroll quite the same way again.
Setting Parameters like Context Length and Temperature in Ollama Models
2 mins
This article provides a concise, practical overview of modifying Ollama models by editing or creating a MODELFILE. It explains how to adjust fundamental parameters such as num_ctx (context length), temperature, and sampling settings, as well as how to modify prompts, templates, and integrate LoRA adapters. The focus is on concrete configuration steps rather than retraining, making it a straightforward reference for those who need to adapt model behavior for specific workloads. The examples demonstrate increasing the context length for code-oriented models, altering generation variability, and attaching domain-specific LoRA modules. The discussion also addresses the technical implications of these changes, including VRAM requirements, inference speed, and the limitations imposed by a modelโs original training context.
