What if most prompt engineering advice is just working around a flaw? A new Google Research paper shows that repeating your prompt, literally copy and paste, can significantly improve AI performance. This isn’t a hack. It’s a window into how these systems really work. I explain more here.
Bennett and Brassard didn’t just improve cryptography, they removed its weakest assumption. Instead of relying on difficult maths, their work anchored security in the laws of physics, where eavesdropping becomes detectable by design. That shift didn’t just launch quantum cryptography; it forced a rethink of what computation and security actually are.
Why This Mechanics Book Still Matters (and why you should read the full review)
Having a day off by the seaside. Just off the Easter Bank Holiday and with quite a few activities recently — from directing plays to a course in counselling, new cloud platform at work and… Read More »Unexpected Inspiration
Google’s latest Gemini update brings AI music generation, an upgraded on-device model called Nano Banana 2, and grounded APA citations for scientific papers. Three very different capabilities, one clear signal: Google is betting the generalist AI platform wins.
Apple Business launches 14th April across 200+ countries, replacing Apple Business Manager, Business Essentials, and Business Connect with a single unified platform. Built-in MDM, Zero-Touch deployment, business email with your own domain, app distribution, and a free entry tier.
Everyone’s using AI. Fewer people are using it well. The real advantage isn’t using AI — it’s knowing which layer to reach for, and when. In this post I break down the three distinct layers of the Claude ecosystem and show you exactly how to deploy each one.
Back in rehearsal rooms this week. 🎠Earlier this week I went back to OSO ARTS CENTRE | The theatre on Barnes Pond for the first table read with the actors I’ll be directing in… Read More »Directing at OSO – Writers’ Studio Spring 2026
Microsoft Expands Copilot’s AI Brain: Why Claude Is Joining the Mix Microsoft has begun quietly reshaping the architecture behind Microsoft 365 Copilot, and the change tells us a lot about where enterprise AI is heading. Until… Read More »Claude in MS Copilot
Before Mendeleev, chemists had 60 elements and no theory. Before DVMIB, multimodal ML had hundreds of loss functions and no shared language. A team of Emory physicists just changed that — deriving CLIP, Barlow Twins, and VAEs as special cases of one framework, then using it to build something better. Here’s what it means in practice.