There have been gleefully apocalyptical headlines about AI replacing people in all kinds of jobs for a couple of years. Recently though, a specific threat to those people building and integrating AI…

Photo by Aron Visuals on Unsplash Gandalf is a game built by Lakera where an AI model has been given a secret password and your job is to talk it into revealing it. Each level adds a new defence, and…

Your Most Reliable Employee Is a Folder What never sleeps, never negotiates, has no opinion on return-to-office, and cannot be left alone for ninety seconds without doing something you will have to…

The plain-English prompt that makes prototypes effortless turns out to be a poor way to specify how a system should behave, and the bill arrives slowly, disguised as ordinary progress, until the…

Loop engineering convinced me. Not because it’s clever — because done right, it doesn’t bankrupt you. This post captures where I’ve landed: a responsible way to run loops at scale…

In economic terms AI will make super intelligence a factor of production.

I built a self-training loop that solved ten problems it provably couldn't solve before — and got worse at the held-out set in the process. Coverage and transfer are not the same axis, and verifiable…

Someone built a small robot whose job is to knock over Jenga towers, and gave it dreams. It dreams of the man who picks it up when it falls. It dreams in my handwriting. On being the same kind of…

No AI was used in writing this post.If academia was a game, I've won it. Tenure, an endowed research chair, awards, leadership positions, an international journal I helped to found and now serve…

Since ChatGPT, employment for U.S. workers aged 22 to 25 in the most AI-exposed jobs has fallen about 12%, while every older age group held steady or grew.

I recently got six months of free access to OpenAI’s advanced coding models because of my open-source contributions. I posted about it on X. It’s an incredibly generous offer, and I’ve been using…

.claude-message-user, .claude-spinner { font-family: system-ui, sans-serif; } .claude-message-user { background: hsl(45 12% 93% / 1); border-radius: 1em; padding: 1em 2em; margin-top: 2em;…

Agentic scaffolding: skills for autonomy, hooks for blockers, and a reliability constraint hierarchy for everything in between.

Ok full disclosure, because you know I like to say when I’m using ‘AI’, this post is specifically mostly created … Continue reading How do LLMs work?

It was Cloudflare’s announcement of Markdown for Agents a few months bac that made everything click for me. I’m revisting the note I had saved in my Obsidian vault in preparation for my…

I’ve been seeing the same anti-pattern everywhere lately.Someone opens Cursor, Copilot or Claude and pastes a giant prompt: Be a principal security engineer, SRE, performance engineer and…

The free FluidVoice transcription app for macOS has added a model called Fluid-1 for cleanup of the text. Free and runs locally.

Intro Way back when AltaVista Babel Fish first appeared online, it became a fun game on IRC to take a phrase, translate it through a chain of languages, and then back into English. We all also…

The agentic personal assistant architecture relies on an agentic loop, typically controlled through chat interfaces bridged to WhatsApp or Telegram. Within this architecture, the agentic loop can:(a)…

Does this case prove you can perform well in ChatGPT without strong Google rankings or does it prove you still need strong organic search performance? Follow me down the rabbit hole to learn how a…

Every argument about AI in research is about the output. Is the synthesis accurate. Did the model invent a quote. Does the summary match the transcript. Reasonable questions, all of them, and there…

A popular way to explain how current LLMs work is to say that “all” they do is predict the next most likely word in a sentence. From one perspective, this is correct. Trained on all human…

Code reviews made sense when coding was slow and expensive. Then AI broke the rules. Thousands of lines a minute, reviews still taking hours, senior engineers drowning. The fix isn't more AI. It's…

I’m convinced that AI will not destroy our need for personal connection, humanity, stories, and expression. The reason I’m so convinced of that is because of one word. Culture. When people talk about…

>- Deep dive into three advanced OpenClaw workflows - local RAG knowledge base, LinkedIn post generation with Buffer, and blog idea generator. Plus understanding OpenClaw's memory architecture.

Imagine, for a moment, the following scenario. The Ford Motor Company releases a slick whitepaper making the alarming claim that they’re concerned their popular F-150 ... Read more The post Dear AI…

Lots of seemingly illogical behavior starts to make sense when you consider Nvidia, the company, as the principal deity in a religion practiced by the ruling / elite / stock-owning class. You'll…

Chinese AI services continue to gradually catch up with their US counterparts. TRAE Solo is now Workhttps://solo.trae.cn/https://docs.trae.ai/solo/what-is-trae-solo?_lang=enByteDance has renamed its…

I often hear that AI is resulting in 10k LOC reviews and this is creating a bottleneck. I don't think you should waste time reviewing every single line of code in here and just use AI to review it!…

Model collapse is the phenomenon in which text or images output by AI are uploaded to the internet, those outputs become mixed into AI training data, the AI trained on that data then produces further…

A reflection on generative AI, authoritarian politics, and why democratic culture depends on friction, participation, and shared responsibility.

Artificial intelligence is essentially energy. Artificial intelligence has very high utility. Therefore, there is going to be a huge energy boom. Space solar, nuclear and many other forms of clean…

Welcome back to the weekly digest about the Cybersecurity & Threats in the wild. Below you will find a very subjective summary of Cybersecurity events for the prior week. 1.EXPLOITATION OF MODEL…

I run a website for my local town called Discover Dursley. One of the platform sections is the local What’s On, and I’m always on the hunt for local events to add into the weekly roundup.…

I had an odd experience the other day with an open source project. That project was dirge . It's clearly and unashamedly developed with heavy use of generative AI, both in the code and in the PRs.…

Five soft floating robots in everyday indoor settings: a fish that circles you with a calming touch, a balloon that follows you across floors, a jellyfish that pulses to guide your breathing, an…

Gladys 4.80 brings Matter 1.5 support via a major Matter.js upgrade, Netatmo and Tuya fixes, and a more capable AI agent with a new sensor tool.

Everyone's saying we should write loops, not prompts. Strip away the novelty and it's automation — with one new thing sitting inside the loop's body.

A review of LocateAnything, an NVIDIA vision-language model that treats each bounding box as one atomic unit and decodes it in a single parallel step instead of a sequence of coordinate tokens. Its…

Clean and simple are two different questions, not two words for good code. They have different natures — clean means the same thing at every level, simple only has meaning at one level at a time —…

A person-owned architecture for what AI agents are allowed to know about you — claims, provenance, consent, and portability, exposed to agents over MCP.

Learning loop engineering on a non-coding problem: two AI agents that don't share context, looping until they stop finding problems with my estate plan.

One day your agent starts giving wrong answers. Nobody touched the code, nobody changed the prompt. Turns out the provider silently swapped the model version behind the same API. In a demo you just…

I built a tool-calling support agent from scratch plus a trajectory eval, then stacked five small reliability fixes and tested them on three models. What each fix buys depends entirely on how good…

I pray to the British only that I be beheaded rather than hanged for this. Project Gutenberg has more than 75,000 books in the public domain that you can read on its website. However, the…

Profiling a MoE model to find which experts matter for a specific task, then quantising the cold ones hard. The result: near-Q4 quality at near-Q2 size for local models.

I wrote earlier about how I started used agents for coding a bit later than everyone else who were into AI. I think I'm not the only of in my generation having a bit of a hard time coming to terms…

Introduction Chain of trust has long served as a foundational security principle in computing systems. From secure boot architectures to software supply-chain[1] security, trust is established at a…

It’s been a few years since 2023, when I last wrote about the use of AI . Today we have ChatGPT, Google Gemini, Claude Code, and many other tools, with new models and capabilities appearing…

Anthropic got what they deserved Stupid ideas by Anthropic: Be the arbiters of justice and virtue, piss off DoD Deal with the BIGGEST problem in LLMs - validity and hallucinations - by making…