Blog
Practical insights on AI technologies and real-world applications
Claude now writes in invisible ink. Can you wash it out?
Luka Breitig
Anthropic quietly began watermarking everything Claude writes. I went through the announcement, the law that prompted it, and the research on whether a watermark like this can be removed. Most of the advice going around is nonsense, but not all of it.
Read more
Do LLMs really write better Elixir than Python?
A Tencent benchmark ranked Elixir #1 for LLM code generation. As an Elixir developer, I dug into the data — the truth is more nuanced than the headline, but the structural advantages are real.
Read more
From chatbot to agent: a step-by-step mental model for understanding agentic AI
Moving from conversational mimicry to autonomous execution: understand how agentic AI systems pursue complex goals, use tools, and adapt to changing environments through a continuous reasoning loop.
Read more
When your AI turns against you: defending public chatbots from prompt injection
Users turned Bild.de's public chatbot against the brand. Here are the prompt-injection defences every team running a public-facing AI should put in place.
Read more