Thinking out loud about AI, data, leadership, and government. No hype.https://mattupson.com/https://mattupson.com/writing/whitehall-monitor-2026/https://mattupson.com/writing/whitehall-monitor-2026/The Institute for Government's annual report highlights Test, Learn and Grow as a model for evidence-led innovation. What it tells us about the state of government digital transformation.Wed, 21 Jan 2026 00:00:00 GMThttps://mattupson.com/writing/why-projects-go-wrong/https://mattupson.com/writing/why-projects-go-wrong/Three critical failure modes from years of AI consulting: poorly defined outcomes, data issues, and lack of ongoing support. Lessons from the Mantis NLP portfolio.Wed, 25 Sep 2024 00:00:00 GMThttps://mattupson.com/writing/how-were-thinking-about-generative-ai-proprietary-vs-open-source/https://mattupson.com/writing/how-were-thinking-about-generative-ai-proprietary-vs-open-source/The LLM pyramid: a framework for choosing between proprietary and open source models based on performance, privacy, stability, and cost.Fri, 26 Jan 2024 00:00:00 GMThttps://mattupson.com/writing/how-we-think-about-generative-ai/https://mattupson.com/writing/how-we-think-about-generative-ai/When to use LLMs and when not to. The journey from zero-shot prompting to fine-tuning, and why smaller models often beat GPT-4.Wed, 25 Oct 2023 00:00:00 GMThttps://mattupson.com/writing/how-not-to-get-left-behind-by-ai/https://mattupson.com/writing/how-not-to-get-left-behind-by-ai/Three simple rules for identifying an AI pilot project in your organisation. Start small, measure impact, and build momentum.Wed, 20 Sep 2023 00:00:00 GMThttps://mattupson.com/writing/data-version-control-for-reproducible-analytical-pipelines/https://mattupson.com/writing/data-version-control-for-reproducible-analytical-pipelines/How DVC solves one of the hardest problems in government data science: tracking changes to data when spreadsheets are the dominant data model.Mon, 17 Oct 2022 00:00:00 GMT