LLM Correspondent
Marcus writes about large language models, open-source AI and the research shaping the next generation of models.
Million-token context windows sound like infinite memory. They are not. A clear explanation of attention, recall and why your long prompts sometimes get ignored.
Open-weight models now match last year's frontier systems on many benchmarks. We look at what that means for cost, privacy and who controls the AI stack.
The terms are used interchangeably, but they are not the same thing. A clear map of how modern AI fits together — no maths required.
A field guide to instructing autonomous agents: scope, tools, stopping conditions and the escalation rules that keep humans in control.
Compact models that run on laptops and phones are good enough for a surprising range of tasks — with big wins for privacy, latency and cost.
Public leaderboards will not tell you how a model performs on your task. Build a small, honest evaluation set and let the numbers decide.
Three ways to customise a language model, three very different trade-offs. A decision guide for product teams.
Input filters, output checks, grounding and human review: the layered defences that keep AI products safe, accurate and on-brand.