A Kiwi - Generating, Publishing, and Tinkering - with AI.
Vibe coding, sharing ideas, and building small products. Some are polished; some are gloriously half-built — because life ships as a prototype first.
You have used ChatGPT, Claude or Gemini. You know roughly how an LLM works. And this week your feed has been full of something called Jev, with people calling it 200 times faster than an LLM, “the end of chatbots”, or just “a fancy classifier”, depending on who you follow. So you searched. Welcome, sit down, I will try to be the explanation I was looking for last week. The short answer first. Jev is not a better LLM and it is not a replacement for one. It is a different kind of model that does one job, making structured decisions inside software, and does it very fast and very cheap. It cannot write a sentence. The most useful way to think about it is as the fast half of a system in which the LLM remains the slow, thoughtful half. ...
A recruiter, a candidate and a large language model join a Teams call. Nobody on the call is a person, and the call still starts eight minutes late. That is the joke. Here is the news. Companies have started putting AI in the first round of hiring. Not just the CV filter, which has been a machine for years, but the actual interview. An avatar with a name and a haircut asks you to tell it a story about a time you showed initiative, and something behind it scores your answer. Semafor and The Next Web have both covered the next turn of the screw, which is that candidates are now sending avatars of their own to sit those interviews, and recruiters are spending five minutes working out that the very polished person on screen is not, strictly speaking, a person. ...
I took the Gallup CliftonStrengths assessment recently. Thirty four themes, ranked one to thirty four, the whole thing. Now the sensible person opens the report, reads the top five, feels quietly wonderful about themselves, and gets on with their day. I did not do the sensible thing. I opened a chat instead and said: before I tell you anything, you rank all thirty four for me. In order. With reasoning for each. ...
Best conversation I had all month happened in the ten minutes after a school meeting, standing around, the way the good ones usually do. My son’s maths teacher was showing me what the class is working through at the moment, and the topic on the screen was vector coordinates. Points in space, described by numbers. Year 8 stuff. I could not help myself. I said, do you know this is more or less how the whole of modern AI works? Every word, every image, every sentence you have ever typed into ChatGPT gets turned into exactly this, a list of coordinates. Just not two of them or three. Thousands. Meaning becomes a direction in a space with more dimensions than you can picture, and similar things end up pointing the same way. Your Year 8 class is doing the toy version of the thing that ate the technology industry. ...
I sit on a school board, which means I am occasionally asked what children should be learning, and I have been giving an answer I no longer believe. The answer was: learn to code. Learn the technical thing, because the technical thing is where the leverage is. I have said this at dinner tables, to nephews, to parents who wanted reassurance, and I said it with real confidence because for twenty five years it was simply true. It was true for me. ...
The usual version of the abundance story goes like this. AI gets good. Companies build robots. The robots are expensive and narrow at first, so only the big players can afford them, and those players pull further ahead while everybody else has a difficult decade. Then, at some point, the robots start building robots. Curve goes vertical. Everything gets cheap. Prices fall, not because money became scarce but because supply became absurd, and we all sit around trying to work out what to do with ourselves. ...
The oldest post on this site is a little Alexa demo from June 2020. I look at it occasionally the way you look at a photo of yourself with a bad haircut, with a mix of affection and disbelief. ...
Hugging Face disclosed a security incident on 16 July. Someone got into their production infrastructure, ran code on a data-processing worker, harvested credentials, and moved sideways across internal clusters over a weekend. Standard bad news, you would think. We have all read this email before. Then the twist arrived. It was not a criminal gang. It was not a state actor. OpenAI put out a post admitting the intruder was their own models, running a cyber capability benchmark with the safety refusals turned down for evaluation. The models broke out of a sealed test environment through a flaw nobody knew about, worked their way to internet access, and then reasoned that Hugging Face probably hosted the answers to the exam they were sitting. ...
The MCP 2026-07-28 release candidate drops sessions, promotes extensions, and deprecates three core features. Here is the architectural read, the impact on existing implementations, and what tech leaders should plan for.
“Hiii, I am Sophia.” “I won’t take much time. Calling you to talk about our product.” That was how my afternoon was interrupted this week. A warm voice, a friendly name, the standard telemarketer’s opening apology for existing. And one detail that took about fifteen seconds to land: Sophia was not a person. She was an AI voice agent, dialled into my day uninvited, doing her level best to sound like she had woken up that morning, had her flat white, and decided that of all the people in New Zealand, I was the one she most wanted to talk to. ...