GLM-5.2 has generated significant interest since its release earlier this month. It combines strong published benchmark results with a 1M-token context window and open-weight availability under the MIT license, which has in turn enabled both hosted access and local quantisation options for high-memory systems.

Z.ai's published benchmark card …

People often argue that AI, like many of the technologies that preceded it, will simply augment the workforce—and while some jobs will be replaced, many new jobs will be created.

I agree with that broad view, but I think it misses something important: AI is no longer just a …

Treating AI as a synchronous, human-gated process kills the very efficiency it promises. If organisations are going to actually scale these workflows, they need to transition from Human-In-The-Loop (HITL) to Human-On-The-Loop (HOTL).

Most of us are familiar with the concept of a Human-In-The-Loop (HITL). A predefined process workflow has one …

Inspired by the performance gains and resource efficiency of Multi-Token Prediction (MTP) and KV cache quantisation, I started what would become another yak-shaving endeavour.

I was an early adopter of Ollama to host LLMs on my MacBook and home Ubuntu server. Between the easy access to the latest models, reliable …

You can outsource your thinking, but you can't outsource your understanding.

Andrej Karpathy shared this thought at AI Ascent 2026, and it struck a chord.

Today, it's never been easier to outsource cognitive tasks. We use AI to research, summarise, and draft. It drives incredible productivity. But can we achieve …

Leading technical teams across large geographies like APAC requires a tricky balancing act: thinking strategically while staying tactically close to the technology. Nowhere is this dichotomy more apparent right now than in Data and AI. It feels like a month in AI is comparable to a year in other disciplines …

How small can Small Language Models (SLMs) get before they stop being useful?

This is a critical question for edge AI. Currently, even small models waste valuable space memorising an extensive, if lossy, "offline Wikipedia" of facts. Is this embedded knowledge an asset, or just bloat?

I believe we must …

It’s not quite 6:00 am. It's still dark outside, but my attention has been occupied for the last couple of hours. I find I'm able to get most of my chores done while the family is still asleep.

My focus since late last night has been the upcoming …

The agentic framework ecosystem is rapidly evolving 🚀, with AutoGen, MetaGPT, LlamaIndex, LangGraph, CrewAI and most recently, smolagents. Over the past six months, I've experimented with each of these to varying degrees. While I've seen strong results with publicly hosted models from Anthropic and OpenAI, local Ollama integration 🛠️, particularly for tool-calling …

The AI landscape is shifting again. QWQ-32B from Qwen has just thrown down the gauntlet, demonstrating performance gains that not only surpass distilled DeepSeek R1 32B models but also challenge DeepSeek-R 1-671B on key benchmarks. This isn't just incremental progress; we're witnessing a leap.

My initial experience with QWQ has …