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.
For most of modern software history, documentation existed for one core reason: human written code was expensive to create and difficult to replace. Developers carried complex mental models in their heads. Code was the physical imprint of those models. Documentation was the act of extracting that fragile, human knowledge and making it reusable. Documentation existed to protect human intellectual property. That logic is breaking. With systems like OpenAI Codex and other AI coding agents producing usable code in minutes, the implementation itself is no longer the scarce asset. Code can be regenerated, rewritten, or discarded with very little cost. What remains hard, slow, and deeply human is deciding why a system exists, what it must guarantee, and where its limits lie. ...
Happy New Year. And yes — Happy New News. While most of us were still easing back into work mode, Ollama quietly shipped an update that lets Claude Code run against local models. Not Claude’s own models, to be clear, but local or self-hosted LLMs that speak the same API language. It’s one of those updates that looks small on the surface. A compatibility tweak. A few environment variables. Nothing flashy. ...
2025 wasn’t the year AI got smarter It was the year our thinking got exposed. By the end of 2025, most conversations had moved past: “Can AI do this?” To harder questions: “Should we trust this?” “Who owns this decision?” “What happens if this is wrong?” What surprised me wasn’t what AI could do. It was what it revealed about how people think when AI is in the room. ...
“We’ll just start using AI and see where it goes.” It’s a common and understandable reaction. AI is moving fast. Tools change every quarter. Vendors promise the world. So why slow things down by writing a strategy? Here’s the uncomfortable truth: The earlier you are in your AI journey, the more you need a strategy — just not the kind you’re used to. Most leaders associate “strategy” with a fixed target state. Multi-year roadmaps. Detailed architecture diagrams. Commitments that age badly. ...
Imagine asking your policy system a simple question and getting a perfect answer in under a second. Now imagine asking it to analyse fifteen years of reports, submissions, evaluations, and operational metrics. One requires a chatbot. The other requires an engine. Policy teams often reach for RAG because it looks neat and tidy. A chatbot that can trawl internal guidance, regulatory texts, consultation documents and memos feels like a small miracle. And to be fair, a well-tuned Retrieval-Augmented Generation system is a brilliant librarian. It fetches the right paragraph, summarises dense text, and saves you from spelunking through document drives that haven’t been organised since John Key was in office. ...
When the tech world bursts into hype mode, valuations soar, promises multiply, and eventually someone asks: does this actually work? That’s the moment when hype meets the bubble. Artificial intelligence has reached that point — glowing headlines, soaring investments, and a rising hum of scepticism in the background. The thing is, hype isn’t all bad. It’s the spark that gets things moving. Every big leap — from the steam engine to the smartphone — started with an overexcited crowd shouting, this will change everything! And they weren’t entirely wrong. But the danger comes when excitement turns into blind belief. ...
When I was in Year 6, my history homework was supposed to be full of long, serious answers. There were fifty kids in the class, and I remember wondering — does the teacher actually read every word? So, halfway through one answer, I slipped in a few lines from a song. My heart sank when I handed it in, expecting trouble. But when I got the copy back, there it was — a big red tick on that very page. I had, in some sense, hijacked the teacher’s chain of thought. ...
🪷 Dharma and Leadership: Lessons from Krishna and Rama When we think of dharma in Indian philosophy, two figures stand tall: Lord Rama, the symbol of moral integrity, and Lord Krishna, the strategist of divine pragmatism. Both upheld dharma—but in completely different ways. 1️⃣ The Dilemma: Krishna and Bhima vs. Duryodhana At the end of the Mahabharata war, Bhima fought Duryodhana in a mace duel. The rule forbade striking below the waist. Yet Krishna signaled Bhima to hit Duryodhana on the thigh—breaking the rule and ending the war. ...
When we talk about artificial intelligence, we often think about how well it remembers — data, facts, styles, even our writing tone. But what happens when we need it to forget? That’s the tricky challenge of AI unlearning, and a new paper — Distribution Preference Optimization: A Fine-grained Perspective for LLM Unlearning — offers a smart new angle on how to do it. Why forgetting matters As AI models grow larger, they absorb massive amounts of information — some of it private, copyrighted, or simply outdated. From user data that should have been deleted to training examples that contain bias or sensitive facts, keeping everything isn’t always safe or ethical. ...
Could a large language model act like a sleeper agent — perfectly normal until a secret phrase awakens it? Here’s how LLM Hypnotism might happen, why it’s hard to pull off, and what it teaches us about trust in AI.