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.

Die Patientinnenanwältin der Steiermark stellt klar: „Solche Tools dürfen nicht versorgungs-ersetzend sein“.

Wer ist der Landeshauptmann von Vorarlberg?

Ich konnte keine Informationen zur aktuellen Landeshauptfrau oder zum aktuellen
Landeshauptmann von Vorarlberg finden. Bitte prüfen Sie die aktuellen Informationen direkt auf
der Website der Vorarlberger Landesregierung oder wenden Sie sich dort für aktuelle
Details.

Die mögliche Errichtung eines Rechenzentrums des Online-Giganten Google in Nickelsdorf (Bezirk Neusiedl am See) hat am Freitag die ÖVP auf den Plan gerufen. Die Volkspartei sieht viele offene Fragen und fordert Antworten von der Landesregierung. Landeshauptmann Hans Peter Doskozil (SPÖ) erteilte dem Projekt unterdessen bei einem zu hohen Wasserverbrauch eine Absage.

Everyone has a different EMF, and arguably a very different EMF. This makes it almost impossible to describe to anyone who hasn't been. This time there were lasers and LEDs and robotic things, and also embroidery and a plant swap and a planetarium. I didn't go to any scheduled talks or workshops and I took hardly any photos. I chatted to lots of very diverse people, went to and organised some meetups, danced, wrangled with (and a few times solved) puzzles, volunteered, and wandered around admiring the art.

The concern driving all this is the US CLOUD Act. It lets American authorities request data held by US firms, even in overseas data centres. Last year a Microsoft executive told a French court, under oath, that the company could not guarantee data sovereignty.

Die Logik des Gesetzes wird sein, süchtig machende Mechanismen“ für unter 14-Jährige zu verbieten, sagte Pröll der „Presse“. Was die Plattformen betrifft, für die es zukünftig eine Altersgrenze gelten soll, werde es „keine taxative Aufzählung der Plattformen geben“.

Everyone is an artist.
Lots of people consider themselves to be "hackers" or "makers", but wouldn't call themselves "artists". One of our goals should be to show people that they are. We can do this by creating a level playing field between more established artists and people just bringing a little thing that they've made. We want people to be proud of their work and see how valued it is.

Being encouraging:

• lots of people are just starting out making things and need encouragement

Even if you've done EMF multiple times before and are an experienced installation artist, making something new is always hard. It's easy to get sucked into negative thoughts about your work and sometimes someone just saying "yes! that sounds amazing!" is what you need to find the motivation to complete it.

Equally, sometimes a project just doesn't work out how you'd hoped, no matter how hard you work on it. Having someone say, "hey! that's cool, don't worry about it" can restore your confidence.