Self-generated prompt injections in compaction summaries. In Our framework for reporting model misalignment OpenAI provide "six reports on unexpected or concerning model behavior we’ve observed in the last six months". This one here is my favorite: they...

Unlike standard RAG, which is “stateless” (forgetting everything after each query), an LLM Wiki is cumulative: Compilation over Search: Knowledge is “compiled” once and kept up to date. When a new document is added, the LLM doesn’t just index it; it updates existing...

Fluency is important but it’s not enough. Being AI-fluent in 2026 is like being email-fluent in 2005. What protects you is learning how to ask the right questions to guide your work with AI. Another key is having accountability for the outcome. That builds...

Go to Source

For decades, companies have bought HR platforms to solve HR problems.  Need to manage performance? Buy a platform. Need to automate recruiting? Buy a platform. Need to answer employee questions? Buy a platform.  But what happens when HR leaders can build AI agents...

Law firm Latham & Watkins has bought its own Nvidia GPU servers and started customising AI models, an expensive and technically demanding set-up that shows how much the technology is changing Big Law’s business model.  Latham, the US’s second-largest law firm with...