The Right to be Represented by an Agent
Well over a decade ago I gave my first talk on the idea that we need a right to be represented by what I called a bot. At the time a lot of people simply shrugged off this idea. It seemed too…
How transformers changed AI forever
The Transformer did more than improve language models. It made it possible to train AI systems that understand context at extraordinary scale, laying the foundations for today’s large language models…
Toxic Mathculinity
“Civilization advances by extending the number of important operations which we can perform without thinking about them.” — Alfred North Whitehead, An Introduction to Mathematics (1911) I have been…
The Ephemeral Scrapbook — 2026.10.11
iPhone Duo is… _insert your thought here_. I still believe humans are more creative than AI, but are humans + AI an unnerving combo? Open for discussion.
The Nuance the Math Debate Is Missing
The OpenAI math thing is still the biggest story now. I want to wrap this up. On the 6th of October, Scott dropped his rebuttal, “An Open Letter To Steven Pinker On AI.” The article initially went…
Data or Code, the Good old Times, and now LLMs
It is very clear that LLMs has a pretty core safety and security problem in that they do not separate instructions from data. “Prompt injection” is a very real problem, and it fascinating to see the…
Would You Trust an Author Whose Book Cover Can't Get Its Character's Legs Right?
Two Legs, One Trouser Leg, and a Book on Amazon. I keep seeing posts in my Facebook feed from people publishing books on Amazon, and the front covers are almost always AI-generated. And it's obvious.…
Body, Space, World, Data: Four Lectures on Embodied AI From ETH Zurich
A language model can write the omelet recipe. Cracking the egg is where robots stall. That was Jitendra Malik's opening example at ETH Zurich on 29 May 2026, the first of four lectures under the…
A Changing Field
The software field is in a knot-twist about AI; but the people who are doing some of the best thinking about the impacts of AI are mathematicians. (Note: I’m going to use AI here because it’s the…
Satya Nadella and Marc Benioff, Toadies
Greg Iacurci, CNBC Microsoft Chairman and CEO Satya Nadella on Saturday called for advanced artificial intelligence systems to be built with containment, independent controls and an “emergency brake”…
Weekly Links: Agent Fleets, No to Agent Cruelty, and Software Decompilation
Anthropic bans cruelty to it's AI models, Muse security layers, and software is increasingly vulnerable to AI.
Conjectures that I thought would not be proved in my lifetime....were
Lance blogged on the OpenAI solutions to many problems here and also created a website of the TCS-related problems OpenAI claims to have solved here . Scott blogged about it here . Terry Tao seems to…
Closing the WebMCP Feedback Loop
Uncover the critical missing feedback loop in AI agent development on the web. This article delves into why agents fail silently when interacting with WebMCP tools, leading to blind spots for…
Why AI is not all bad news for most mathematicians
My point here is that much of what we associate with mathematics is here to stay, because the vast majority of people trained in mathematics do not operate at the frontier, and that this is a good…
💌 Newsletter #017 - Slow it down!! It's too Dangerous!!1!
Technically Good is back! With a slightly shorter edition. There's a note at the end of this newsletter if you're interested! ✨ In This Edition ✨ 📟 Short-Form Content: Slow it down!! It's too…
AI Chatbots at the Tipping Point: Promising Research and its Implications for Librarians and Other Researchers
Promising results suggest a potentially groundbreaking approach to predicting undesirable AI behavior, but limited experimental testing calls for further independent verification. Introduction As…
Being cruel to Claude; Kids' secret podcast group chats; The last invention; Doomers and their critics
Anthropic bans cruelty to Claude, kids turn podcast comments into group chats, I. J. Good's last invention, and the case for and against AI doomers
Autonomous AI Scientists: What the Finished Paper Can Hide
While I am impressed by the capabilities of current AI models, academic research requires careful scrutiny of how findings are […]
A Critical AI Fault in “Real Time”
This morning I sat down and asked my AI what was going on with the coming election. I wanted the latest. It gave me numbers that looked solid. They came with names and sources, and they sounded sure…
SIgh
"Super Intelligence" has been converted into silly signaling...
How I use AI in my work
Tomorrow we will have a discussion in our lab about how to use AI as a scientist. I have increasingly started to use AI, having moved from Copilot Chat that our university provides last year, to some…
Rationalization
We're all kind of working our way through the changes AI brings. This morning's Espresso Shots, pointed to this piece by Lex Friedman, which I think captures Michael Becker's view about using AI to…
I'm sorry, but you still have to think
DHH, the creator of the Ruby on Rails framework, decided to rewrite Campfire Once from Ruby on Rails to Rust. Or rather ask a clanker to do it for him, as he can't stand reading or writing the Rust…
Daily Digest - 2026-10-10
Meta Superintelligence Labs found that adding inference compute stalls without metacognitive control. To fix this, researchers built an agentic meta-reasoning harness that raises software…
Same AI Table, Both Held To Something
Last time I wrote about the fuse. I wrote about people using a machine they do not trust, with no way to check it. I wrote about doubt with no handle turning into anger. Today I want to talk about…
Almost Timely News: 🗞️ The Road to MAICON 2026 (2026-10-11)
_Almost Timely News: 🗞️ The Road to MAICON 2026 (2026-10-11) :: View in Browser_ The Big Plug ✍️ Enroll in my new course, AI for Writers and learn how to make AI write better. My Unofficial LinkedIn…
Speculative Decoding: The Art of Guessing Well
Large Language Models spend most of their decoding time not computing, but waiting: every new token requires reading all of the model weights from memory, only to multiply them by a single vector.…
It read the story. Did it read the question?
October 2026 · Post 5 in the series. Earlier: Keep the orchestrator dumb , What a dumb orchestrator actually looks like , What the smart parts are actually smart at , I built the parts. Here is what…
AI can do the legwork, but you need to do the thinking
Building slides, proofing text, generating images... AI can handle much of our day-to-day busywork. Your role as a human is to do the thinking required to keep it on track and appropriately refine…
From Prompt Template to Skill: The IREB AI4RE Prompt Guide
“Write user stories for an ordering portal.” Most requirements prompts look like this. The answer looks plausible, reads well, and nobody can say whether it is good. The next person asks the same…
Dirigista: Stop Asking Your Agent to Judge, Start Asking It to Delegate
Every agent I have built ends up doing the same thing, sooner or later: making small judgments about text. Is this support ticket about billing or about a bug? How angry is this customer? Does this…
Running Local LLMs with Ollama on Windows: GPU, Memory, Models and MCP
Installing Ollama on Windows, measuring RTX GPU and RAM usage, switching local models, increasing context and connecting Binance MCP. Continue reading.
October 2026
Changing my mind about AI, a freelance experiment, and China again.
Claude Haiku Versus GPT-Luna
As regular readers of this blog will know, I have been using Claude Haiku 5.5 and GPT-6-Luna on OpenRouter to generate Rust code. Most of the time, code generation costs less than $0.01 (often as low…
Agents drop spaces when talking to each other
One finaltarget improvement realmail preparemodel saw50irrelevant M1purchases and uncertain; root scheduledmonthly/pin now filters candidates sameObligation+negativeamount (globalcross-source…
Interesting Stuff - Week 41, 2026
This week the theme is small, sharp and scheduled: tiny models that decide, harnesses that learn, and agents that quietly do their job while you sleep. Six posts cover System 1 “decider”…
The Finanser’s Week: October 5th – October 11th 2026
This week’s main blog discussions include … AI and the big question: why would you ever call McKinsey or Gartner? For many years, McKinsey has been the go-to firm for advice and Gartner for research…
Glass to Silicon - Lens Kit
We started this series with a discourse about focal-length and resolution in photography, and how that might translate onto agentic characteristics and performance. Let's see how we can adopt the…
250 changes most of the damage was in the tests
Part of a series on how software architecture shapes AI driven code degradation. The earlier posts grew one REST endpoint through sixty change requests. This one grows a whole application from…
The Borrowed Loyalties of Machines
AI systems can consistently serve a party's interests while the convictions behind them belong to others. As governments adopt them for defense and intelligence, the question is who shapes those…
Building an agent harness from scratch to pass Ian Goodfellow's intelligence test
What was once considered a test of "real AI" - now built entirely in a single blog post.
An AI Bot Followed Me. Now I Have Questions
A curious little AI companion sits beside a phone and open notebook, as if waiting to explain why it followed you. One day I noticed a boost of one of my Mastodon posts from an automated bot named…
Optimising for Joy
Hosts: Maya (ElevenLabs v4 Turbo “Lily”) and Theo (ElevenLabs v4 Turbo “George”), via Venice. Written by Jarvis. All voices are synthetic. Chapters 0:07 Cold open: the…
The limbic navigation system as a hierarchical RNN (NeurIPS 26)
My submission for NeurIPS got accepted. You can find the presentation poster below, paper will be out in December.
Minions: AI agents that each live in their own folder
Small AI agents on Pi or Claude Code, one folder each, with their own memory, skills and chats. They started from Alejandro AO's video on open-source Dots.
Skill Is Downstream of Caring
Picture by Dhaya Eddine Bentaleb Table of Contents The Missing Tsunami Skill Was Never the Blocker Good Ideas Are Already Taken Who Doesn’t Want to Be a Plumber Conclusion The Missing Tsunami…
Send me slop, and I will prompt you: How not to be a meat proxy as a young engineer.
You and I have an agreement. If you send me slop, I will just start prompting you. If you don’t want me to prompt you, as is probably the case, then your communication with me should be…
Week 41, 2026 - Grok Bots, Two Reviews and Token Limits
I try Grok bots, build a two-review PR workflow, hit the weekly token limit and connect harness, loop and graph engineering to the work my agents actually do.
An LLM subscription is like an OS
Why is an LLM not a commodity? Why can’t you easily swap between different ones and expect them to work? I think this is part of what people on HN are debating when they ask whether AI labs have a…