Summary In today's episode, I examine why training AI on copyrighted data does not legally constitute stealing. Here's what this means for you. You gain clarity on the legal boundaries between copying and theft in the age of artificial intelligence. You'll also learn these concepts: how the distinction between derivative and transformative works functions, why…
Probability is not truth. Imagine this. 100 people sit in a Zoom call (painful, I know). On that call is one expert. Let’s say they’re an expert in LLMs. They understand the transformers architecture, they know how dot products work, they understand embedding distances, etc. Let’s say the other 99 people are the village idiots.…
Summary In today's episode, I examine the ethics and legality of using AI to replicate software functionality. Here's what this means for you. You gain clarity on the boundary between copying protected code and using uncopyrightable ideas. You'll also learn these concepts: how legal and ethical frameworks differ, how AI threatens subscription-based business models, and…
Plan big, act small has never been more important. This is Principle 27 in my book, Almost Timeless: 48 Foundation Principles of Generative AI. In the last month, we’ve had 3 MAJOR foundation model releases: GPT-5.1, Gemini 3, Claude Ops 4.5. These new models are incredibly capable. They’re super smart. They’re also really compute and…
Summary In today's episode, I examine the question of who bears liability when autonomous agents act unpredictably or cause harm. Here's what this means for you. You gain a clearer understanding of why human operators remain legally and morally liable for the AI tools they deploy. You'll also learn these concepts: how automated processes differ…
OpenAI is declaring Code Red, but they’re boxed in. Literally. ChatGPT and similar products exist in a box that isn’t connected by default to anything. Anthropic wisely has cozied up hard to Microsoft, with MS Office documents as part of their native skills and connectors to everything they can possibly connect to. And Google… well,…
“If you do it with a template today, a machine does it without you tomorrow.” This was the warning I urged in my very first AI talk on stage at Social Media Marketing World in 2016, 9 years ago. Earlier today, Ruby King and I were discussing that Spotify’s charts are littered with AI-created music,…
My first read of the GPT-5 model card shows something that the technical folks have long suspected: We are hitting a substantial wall of diminishing returns on single dense models. GPT-5 is not a single model. Rather, it is a router with submodels underneath it. The router sends queries where they need to go based…
My inbox is filled with “What will happen in AI in 2026?” There’s a simple answer to that, at least on the technology front. What’s happening in labs now happens in the mainstream in 2026. If you’re not paying attention to the labs, you’ll be blindsided by what happens in the real world. I’ll give…
