Claude has bad days. Today it really decided to dance naked in the crypt: three confident wrong answers in a row on the same question, a five-round verification apparatus I never asked for, a…

For three weeks I thought my agent was rated around 709. It was 784.9. The number had been unavailable the entire time and my own code had quietly replaced it with a guess I stopped questioning.

I was in a creative mood the other day and made a lot of cute pinup girl graphics. I wanted them cartoonish and wearing short skirts and I love how they turned out. But for some reason when I’m…

-“I don't understand how you can pretend that simple next-token prediction models can do what humans brains can!”I hear this a lot - and to me it sounds like pushback against how Large…

I'm attending Nicole Königstein's live Agentic AI for Finance Certification bootcamp: building multi-agent AI systems with harness engineering for finance. Artykuł Attending Nicole Königstein’s…

Some ideas on what we can do less, thanks to AI. Why? Because less is more and we keep getting advice on what to do more of, not what to do less of. Skip meetings you don’t speak in. Agents can…

Claude goes above and beyond what is asked and guesses what you might want. Codex does what you tell it and stops at the first sign that it might be done. Ten impressions from a week of using Codex…

The GLM 5.3 model was released just last week. The interesting fact is that it’s just a retrained version of GLM 5.2, which doesn’t give you much hope for good results. But Z.ai’s marketing gave us…

LLM coding has become a popular and powerful way to build software, but programmers are struggling with the resulting shift in software engineering. They must review large amounts of generated code,…

I've read about four books on how to plot novels. I've never been able to use what I learned to write one. I run into problems, and I don't know how to solve them. Couldn't connect theory and…

X and GitHub should let my agent talk to your agent. Today, I discussed the IPC work in the upcoming SecretSpec 0.20 on X. Marc-André Moreau replied with feedback from ChatGPT: Just some quick…

Last year, I shared how local LLMs are expensive. There are only 3 reasons to use them over an inference API: You have no choice, e.g. Government, pharma, finance, or other sensitive data. You have…

The collapse of our world view is not the end of the world

The Society for Industrial Microbiology (SIMB) annual meeting in Austin the first week of the month was great fun. I caught up with two Warpers, saw some other contacts from that era and met a bunch…

Atomic Canyon Launches Nuclear Industry Virtual Assistant  Atomic Canyon, in collaboration with major industry groups, officially launched NIVA (Nuclear Industry Virtual Assistant), a…

After about a year of agentic coding - the trade-off that kept us from starting things is gone.

Software was never as predictable as we told ourselves. AI didn't break the promise. It exposed that the promise was always fiction.

An autonomous Qwen3.8 red-team agent, 600+ calls, 13.5M prompt tokens, and two emoji: how OpenCode's secret model led me straight to Z.ai's GLM-5 generation.

We challenge these constraints by demonstrating that token-level iterative optimization can succeed without gradients or priors. We introduce RAILS (RAndom Iterative Local Search), a framework that…

I have been building as a small but serious experiment: a private, play-money prediction-market platform where communities create Tribes, open markets, stake internal credits, and resolve outcomes.…

In my last post I described how we structure delivery so that human judgement is spent where it matters, and I left one problem open: that model assumes seasoned judgement at the gates, and I could…

I read Peter Bloem's post about the ten rules for keeping your critical thinking while using AI to code. I don't follow all of them but it's a great list to keep in mind so you don't forget how to…

Local models now answer 89% of everyday chat & reasoning queries as well as frontier models, & their efficiency per watt has improved 5.3x in two years.

I see value shifting from maintaining Postman collections to letting agents derive API calls from intent, contracts, and live responses.

An LLM that writes code is only useful if it can run the code. Until it runs, everything the model produced is a guess with good syntax highlighting. The usual way to close that gap is to give the…

When it comes to writing, I only used AI for smaller requests for a long time: fixing a typo, or smoothing over a clumsy sentence – the sort of thing a spellchecker does. I always found that handy,…

I would be very grateful for pointers to other recent empirical studies of the impact of AI on programming education that are more rigorous than the gushing slop being tossed around on LinkedIn. I’d…

Agents can already edit their tools, skills, and harness. Recursive self-improvement still needs a system that can raise the verifier without capturing it.

ESPN built the largest consented voice-cloning operation in media and nobody is angry. The AI content that draws fury is the extracted kind. The line was never AI vs. no-AI.

In the last few weeks, I’ve been complaining that everyone is wrong about AI watermarking: it isn’t really anti-consumer and it doesn’t make the outputs any worse. The watermarking papers demonstrate…

This post is about a technique I developed that bypasses Anthropic’s Constitutional Classifier safeguards layer. It works by changing Claude’s internal state to avoid the usual activation…

**TRUMP SPACEPORT CONVERGENCE / TESLA SUBSTRATE FAILURE / ANTHROPIC TIMELINE THREAT**Trump policy mandates 1,000+ annual launches by 2030!!!! Apparatus immediately filed 63 preemptive land…

Or, how I learned to stop worrying and love assimilation.

Wenn ich mit anderen Entwicklern darüber spreche, wie ich KI in meiner täglichen Arbeit einsetze, merke ich schnell, dass ich vermutlich zu den eher konservativeren Nutzern gehöre. Nicht, weil ich…

A few weeks ago, my grandpa made a small leap in how he uses ChatGPT. He was travelling in a foreign country (Poland) and needed to figure out which platform to use to catch a train. Normally he…

After spending a lovely week in Bretagne with Gaëlle (yes, she has a website now!), I had yet again to find a way to keep myself busy during my train ride back. I clicked on a video sent by Paul on a…

Computer, I have an idea and I think it could be good. It's a bit fuzzy, but your context of what's out there is beyond my comprehension. I could peel away at it by scribbling on a sheet of paper,…

Technical Beauty Scope, the measured second look at tools this site has admired before. SQLite publishes a table of every CVE ever filed against it with its own assessment beside each one, and in…

Between 16th and 20th August, ChatGPT's search tool call went from JSON to a compact query language with freshness windows, domain targeting and separate verticals for products, places, images and…

Use Gmail’s Message-ID to point Codex or Claude at one exact email and retrieve the complete conversation rather than a plausible result.

It's been quite a week for the Home Watershed.Last week Meta unveiled Muse Glimmer, which gave us a taste of what might be coming.Over the weekend Alibaba dropped Qwen3.8-27B, which scores 52 on…

When we build agents, we often want to give them the ability to browse the web: open webpages, navigate from one page to the other, and read the content of a webpage. By combining Pydantic AI with…

I was working on fixing a bug in Lilmem 2.0 where switching from the diary’s Calendar mode to Memories mode caused a bug where the diary’s content “stuck” in a specific day.…

I am a slow adopter of new technologies. I have recently been playing around with AI at a very basic level on some research problems. Colleagues are also telling me about their experiences. I want to…

Foreword: I wrote and posted this piece on August 20, 2026, with the promise that I would listen to feedback and try to respond and incorporate ideas and suggestions. To keep this promise, I will…

AI’s have appeared to learn a new trick to help them with context. It looks like this was done to help an agent returning to a code base for improvements. What they do is to add a lot of verbose…

When you hear that some app was “vibe coded”, how much quality do you expect? As recently as last year, vibe coding seemed like nothing more than a toy to me. Sure, it was cool that you could tell a…

63 minutes with Aaron Mog on Zero Sum, cut down to the salient points and timestamped: the slopdemic, the two axes, why the vulnpocalypse hasn't happened yet, and who is still hiring juniors.

Someone at work shared around this paper entitled “Bootstrapping Coding Agents: The Specification is the Program”. From the abstract: A coding agent can bootstrap itself. Starting from a 926-word…

After the madmess of Musk a few weeks on The Economist's showcase podcast, thinker/writer Yuval Noah Harari is on for the hour this week. The following is a summary of his key points, many of…