Tag: llm
HODLing gear in 2026
A Month Without Frontier Models
Ineffizient
Mit Effizienz misst man die Ressourcennutzung – also das Verhältnis von Ergebnis zu eingesetzten Mitteln (Zeit, Geld, Material, Personal). Der Ressourcenverbrauch von LLMs ist enorm und mir erscheint es schon länger so, als ob man so etwas wie ein Auto erfunden hätte, aber dummerweise mit einem Benzinverbrauch von 10 Litern pro Kilometer. Solange ich das...
LLMs Just "Fake it `til they make it"
This is what happened when I tried to make a Language Model mimic my knowledge and behaviour for a chat-bot on my portfolio site.
What Is a Harness?
LLMs are engines. Harnesses are everything else—the wheels, brakes, dashboard, GPS—that turn a raw engine into a useful vehicle. First in a series on harnesses for the open knowledge commons.
I built a snark detector and pointed it at myself
Toward Automated Discourse Network Analysis
Discourse Network Analysis has long been limited by the price of expert judgment. Here is a design for automating it at corpus scale without surrendering command of meaning — and FineStructure, the open-source workbench I am building for it.
Language integrated LLMs as an OCaml function
Using a local DeepSeek model as an ordinary OCaml library and building sandboxed agents from simple primitives
OpenSearch Semantic Search
My learnings on OpenSearch semantic searching
OpenSearch Semantic Search
My learnings on OpenSearch semantic searching
Treating LLMs as programming books
Thoughts on an approach for using LLMs effectively for coding without losing engagement and cognitive effort.
Treating LLMs as programming books
Thoughts on an approach for using LLMs effectively for coding without losing engagement and cognitive effort.
Det här med AI eller LLM
Det här med AI eller stora språkmodeller (LLM) som det egentligen handlar om är en fråga med en mängd aspekter och synsätt. På det sociala medium, Mastodon, som jag främst använder är de flesta väldigt negativa till Artificiell intelligens (AI).
yallmap — Yet Another LLM Proxy
Building an Anthropic-native LLM gateway in TypeScript
The social contract of writing
About the value of genuine writing in a world being drowned in slop.
The social contract of writing
About the value of genuine writing in a world being drowned in slop.
Wie kann ChatGPT zur Verbesserung eines Magento Onlineshops verwendet werden?
Dieser Beitrag soll ein paar Anregungen geben, wie man als Shopbetreiber seinen Shop mit ChatGTP verbessern kann. Wozu kann ChatGPT oder ein andere KI Client genutzt werden für Shopverbesserungen? KI Chatclient kann meiner Meinung nach sehr gut als Sparing Partner und Ideengeber liefern. Die KI kann riesige Mengen an Informationen für einen Verarbeiten ohne das...
e554 — SPI vs I
e554 - SPI vs I: Stories and discussion on #LLM #PhoneNumber lookups, #proctors returning to #Princeton, lavish #LEGO, #LOTR and a whole lot more!
Three circles for thinking about LLMs
A Venn diagram for clarifying what's actually at stake when people argue about whether LLMs are intelligent, conscious, or just stochastic parrots.
Running local models on an M4 with 24GB memory
Experiments with getting usable outputs out of local models on a standard Macbook
Running local models on an M4 with 24GB memory
Experiments with getting usable outputs out of local models on a standard Macbook
Collaboration: a Confused Story in Five Graphs
Some graphs about reading and writing on the internet. Less a story than a Rorschach test.
Le Chat – Eine europäische Alternative zu Copilot, ChatGPT und Claude? (Werbung – unbeauftragt, unbezahlt)
Werbung – unbeauftragt & unbezahlt:Aufgrund von Markennennung, Produktdarstellung und Verlinkungen handelt es sich um Werbung, auch wenn ich das Produkt selbst gekauft habe und keine Kooperation besteht. Le Chat (frz. die Katze) ist als Katzenhalter und -liebhaber doch ein sehr ansprechender Name für eine künstliche Intelligenz. Bekannt war mir die KI auch bereits schon länger,...
Wikipedia's traffic drop: more on languages and freshness
Following up on last week's post, I looked at 5,000 "Vital Articles" across eight major-language Wikipedias. Articles about math, physical sciences and tech are waaaay down, while people, geography, and history hold up far better—regardless of which language they're in. Article freshness matters too—but not as much.
Magento Onlineshop Wiederkäufe durch KI/LLMs steigern
In diesem Beitrag soll einmal auf theoretischem Level erläutert werden welche Daten in einem Magento Onlineshop vorhanden sind und wie diese über eine KI-Anbindung "gehoben" werden können. Der Fokus liegt in diesem Beitrag auf den Wiederkäufe. Somit Stammkundschaft stärken. Welche Daten gibt es im Onlineshop? Bestelldaten Der Shop erfasst logischerweise die Bestellungen. Diese Bestellungen sind...
Career articles on Wikipedia: some scary numbers
I took a look at English Wikipedia pageviews for ~4,000 articles about careers. The numbers are grim: the median is down 28% from pre-COVID, with a huge drop in the last year.
Same Agent, Different Score: The Problem With Testing Non-Deterministic AI
Before building tools for my Zork-playing agents, I needed a benchmark I could trust. I ran five local models through fifty playthroughs and discovered that the same model can score 40 or 0 on the same game. Getting honest numbers required three harness versions, structured telemetry, and a loop detector that learned the difference between stuck and thorough.
How to hit your Claude weekly limit so you can go outside and touch grass
A satirical guide to maxing out your Claude weekly limit so you finally go outside and touch grass, featuring sub-agents, MCPs, and max effort.
How to hit your Claude weekly limit so you can go outside and touch grass
A satirical guide to maxing out your Claude weekly limit so you finally go outside and touch grass, featuring sub-agents, MCPs, and max effort.
Stuck in the Maze: Why AI Agents Can't Hold the Map
I had local AI models play Zork, the 1981 text adventure, to study why agents struggle to navigate connected systems. One started responding in Thai. Most scored zero. All got hopelessly stuck in the maze. What broke says a lot about why agents get lost in microservices too.
The model vs the harness
"What model does it use?" is the wrong question. Most of the differences you feel between AI tools come from the harness, not the model.
Stay in the Loop: How I Actually Use Claude Code
How to multi-task Claude Code while staying in the loop, increasing success rate and parallelization.
Connecting the dots for biodiversity action from the NAS/Royal Society Forum
Summary of the Nine Recommendations and Biodiversity Monitoring Standards Framework papers from the NAS/Royal Society US-UK Forum in summer 2025, and how they connect to my work on collective knowledge systems, TESSERA, and evidence synthesis.
Stay in the Loop: How I Actually Use Claude Code
How to multi-task Claude Code while staying in the loop, increasing success rate and parallelization.
Lawyers, Humility, and LLMs
Reviewing a book about a multi-billion-dollar contract bug—and what it means for the profession's arrogant response to LLMs.