Below are some simple methods for exiting vim.

For real vim (and hacking) tips, follow hakluke and tomnomnom on twitter.

Open hardware specialist PINE64 has announced that it is exiting the Linux single-board computer, phone, and tablet market — at least until the bursting of the artificial intelligence (AI) bubble and a drop in sky-high component prices.

"Due to the current DRAM/eMMC shortage, there isn't any plans to continue producing more Linux devices in the near future," PINE64 announced on the community Mastodon account this week. "Whether production continues or not depends on the pricing situation for those components after mid 2027. Devices like the PineNote and PineTab2 are currently estimated to run out of stock in three months time. Just pre-warning everyone in case people start asking when things are going to be in stock again."

HTML has been gobbling up swathes of what used to be JavaScript’s remit. This page lists a bunch of dynamic functionality that we can now achieve with just HTML.

My third post about typography in sci-fi has been gestating for a while now. Indeed, it’s been slowly taking shape – you might say it’s been forming itself inside of me – for really quite some time.

I’m delighted to say that it is now ready to burst forth from my allegorical chest, and to spatter allegorical typographic blood all over your allegorical faces. Welcome to Typeset In The Future: The Alien Edition.

The opening credits for Alien are nothing short of a typographic masterpiece.

Let me start by saying, I vehemently hate AI as it is. It is blight upon our planet. In so many ways.

Because of the "popularity" of AI with managers and CEOs who are, actually looking to replace all of us with a series of if/then statements, AI and LLMS have become the big tech fad these days. This has resulted in a series of repercussions for everyone else:

The AI software/LLMs require massive amounts of extremely powerful hardware, because by design, AI is horribly inefficient. So this has resulted in a "Datacenter Boom" across the planet. Datacenters are essentially just warehouses stuffed with computers. Think of Microsoft Azure, Amazon AWS, Oracle Cloud, your Preferred hosting provider/cloud provider, Private datacenters for large companies, etc.

As you might imagine, the desire to stand up more datacenters has resulted in more hardware being bought up by the big tech conglomerates. Famously, Sam Altman of OpenAI, the company behind ChatGPT bought out 40% of the world's RAM production about a year ago. There are three major RAM manufacturers on the planet: Samsung, Micron, and SK Hynix. Just kidding there's only really two now because Micron has decided that they are no longer selling to the general market and have exclusively gone to AI datacenter sales only. On top of all that, global RAM production capacity has been bought out through 2028. This of course has predictably caused the price of RAM to fucktuple.

I was able to build my nodes for around 30 EUR per device – by cheating a bit, because I had quite a lot of the parts already lying around and we were able to source dirt cheap surplus antennas. Realistically, you can build one such node at about 30–60 EUR, with antenna size and quality being mostly responsible for the 30 EUR leeway.

Am 1. September 2025 beginnt die Zukunft. Ein "Paradigmenwechsel" sollte das vor einem Jahr in Kraft tretende Informationsfreiheitsgesetz (IFG) sein und Österreichs Verwaltung seither so richtig transparent. Und, ja. Auf Basis des neuen "Grundrechts auf Informationen" sind tatsächlich einige wegweisende Entscheidungen gefällt worden. Beispielsweise entschied das Bundesverwaltungsgericht, dass Bundeskanzler Christian Stocker (ÖVP) dem STANDARD Zugang zu fast all seinen Kalendereinträgen als Regierungschef geben muss. (Stocker hat Revision an den Verwaltungsgerichtshof erhoben, dieser hat noch nicht über die Sache entschieden.)

In einer Sache erinnert Österreichs Informationsfreiheit allerdings verdächtig an Deutschland. (Wo die schwarz-rote Koalition übrigens gerade darüber debattiert, das dortige Informationsfreiheitsgesetz einzuschränken.) Denn mehrere österreichische Behörden weigern sich, Anträge nach dem Informationsfreiheitsgesetz per E-Mail entgegenzunehmen. Das führt dazu, dass die Transparenz-Herstellungs-Plattform FragDenStaat.at, worüber interessierte Menschen sehr einfach Anfragen an die Verwaltung stellen können, ab sofort Faxe verschickt.

Das Problem: Der aktuelle Entwurf sieht keine durchgängige menschliche Kontrolle der KI-generierten Bescheide mehr vor. Nur stichprobenartig soll überprüft werden, was die Software da ausspuckt. Das öffnet Tür und Tor für Halluzinationen und rechtswidrige Entscheidungen, warnen Bürgerrechtler.

Im Herbst soll der Verfassungsausschuss des Nationalrats über eine Novelle des Allgemeinen Verwaltungsverfahrensgesetzes (AVG) beraten. Kernstück der Regierungsvorlage ist der neue § 18a, der die Basis für Verwaltung durch die Künstliche Intelligenz bildet.

The GNOME Shell Extension - Quake Terminal enhances your desktop by providing a drop-down interface, inspired by classic Quake games, that can instantly launch your preferred terminal emulator over any workspace.

AI needs a do-over. As it exists right now, I don't give a toss what good it can do, what practical benefits it has once the techbros move on to their next mark. I don't care about any of it at all because AI companies botched the first impressions so hard by telling me consent is a foreign concept and I need to just roll over and submit, or I'll be left behind forever.

It was this brazen disregard for any kind of consent from the get go that did most of the work in turning me against anything AI had to offer.

If you want me to care about AI? Start over. From zero. Consent needs to be a core concept of it. If people don't want to use it, respect that opinion. Do NOT treat every no as a thinly veiled yes. Do not force it on people who don't want it. And if people don't want it, don't scream "you'll like it!" or "it's the future, get with it or get left behind!" at them.

Because that language sounds eerily similar to the type of language used in domestic abuse situations and is that really the way you want to come off? (Regrettably, some of the psychopaths who push this garbage would nod yes with a grin across their face because they're absolutely depraved. These people should not be allowed to be anywhere near the helm of a tech company. But alas, this is 2026.)

Posts to Mastodon and Bluesky completions to the sentence "I hope this email finds you." All content is sourced from Google Books, based on searches for phrases that start with "finds you."

In kommunalen Diskussionen sind die Chancen von Sprachmodell-KI und ihren Agenten seltsam risikobefreit. Dass die Chancen zu höherer Produktivität, mehr Effizienz oder schnellerer Bearbeitung führen, scheint gewiss. Man müsse es nur richtig machen.

Aber wie es generell „richtig“ geht, kann niemand wissen. Use-Cases und Best Practices geben Anregungen, sind aber interkommunal nicht so übertragbar, wie man sich das vorstellt. Die Beharrlichkeit und soziale Intelligenz historisch gewachsener, lokaler Organisationsstrukturen werden unterschätzt.

Die Tirol-Karte wirkt verzerrt, der Landesadler hat zu viele Federn. Eine Grafik auf dem Instagram-Kanal von Innsbrucks Bürgermeister zeigt, wie die Politik KI nutzt und was passiert, wenn das Ergebnis nicht kontrolliert wird.

The US Supreme Court’s Slaughter ruling on Monday, which overturned longstanding precedent called Humphrey’s Executor, is seriously bad news for American companies who handle Europeans’ personal data. That’s because the ruling destroyed the independence of the Federal Trade Commission (FTC), which is an essential part of a deal that enables transatlantic data flows. And now Max Schrems, who blew up that deal’s predecessors, is going in for the kill.

Immer wieder haben zuletzt geplante Rechenzentren für Aufregung gesorgt. Auch in Leopoldsdorf passierte ein solches Bauprojekt kürzlich den Gemeinderat. Wie die „NÖN“ berichteten, ist nun auch ein zweites Projekt geplant. Kritik kommt aus der Nachbargemeinde.

For example, in semantic search, we index a corpus of documents, with each document containing valuable information on a specific topic. Due to the way embedding models work, those documents will need to be chunked, and similarity is determined by chunk-level comparisons to the input query vector. Then, these similar chunks are returned back to the user. By finding an effective chunking strategy, we can ensure our search results accurately capture the essence of the user’s query.

If our chunks are too small or too large, it may lead to imprecise search results or missed opportunities to surface relevant content. As a rule of thumb, if the chunk of text makes sense without the surrounding context to a human, it will make sense to the language model as well. Therefore, finding the optimal chunk size for the documents in the corpus is crucial to ensuring that the search results are accurate and relevant.

These models find semantically similar sentences within one language or across languages:

sentence-transformers/distiluse-base-multilingual-cased-v1: Multilingual knowledge distilled version of multilingual Universal Sentence Encoder. Supports 15 languages: Arabic, Chinese, Dutch, English, French, German, Italian, Korean, Polish, Portuguese, Russian, Spanish, Turkish.

The issue with multilingual BERT (mBERT) as well as with XLM-RoBERTa is that those produce rather bad sentence representation out-of-the-box. Further, the vectors spaces between languages are not aligned, i.e., the sentences with the same content in different languages would be mapped to different locations in the vector space.

In my publication Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation I describe an easy approach to extend sentence embeddings to further languages.

Chien Vu also wrote a nice blog article on this technique: A complete guide to transfer learning from English to other Languages using Sentence Embeddings BERT Models

Characteristics of Sentence Transformer (a.k.a bi-encoder) models:

Calculates a fixed-size vector representation (embedding) given texts, images, audio, or video.

Embedding calculation is often efficient, embedding similarity calculation is very fast.

Applicable for a wide range of tasks, such as semantic textual similarity, semantic search, clustering, classification, paraphrase mining, and more.

Often used as a first step in a two-step retrieval process, where a Cross-Encoder (a.k.a. reranker) model is used to re-rank the top-k results from the bi-encoder.

The two families of similarity

The first family is lexical similarity. This is where local libraries make light work of the problem. If I want to match names, organisations, cities, account references, or noisy OCR output, then normalisation, entity extraction, fuzzy matching, and character n-grams are often fast, cheap, and surprisingly effective.

The second family is semantic similarity. This is where embeddings start to matter. If two phrases are conceptually related but use different wording, fuzzy matching often falls apart. Embeddings give you a vector representation that lets you compare meaning rather than spelling.