Can you build a three layer cake by starting with the third layer? Can you deploy code before writing it? Can you publish an ugly first draft and get great results? One of the biggest flaws I see in AI implementation today is rushing ahead. Lack of patience. Taking shortcuts. Cutting corners. If your idea…
Predictive analytics in marketing is broken. More specifically, predictive analytics at the macro level – like what consumers are going to purchase, how they plan, anticipating demand, etc. is badly broken. And it has been since March 2020. Here’s why. By its very nature, forecasting and predictive analytics rely on historical data. The more stable…
AI struggles with low probability. When you understand the inner workings of AI – namely, that it’s always trying to predict the highest probability sequence, as a prediction engine – then it becomes logically clear that AI struggles with generating low probabilities. Doing so goes against its very nature, the way the system is designed.…
“Garbage in, garbage out.” This truism from the earliest days of computing is still true. In fact, it’s more true than ever; generative AI models are built using massive amounts of data. And not all that data is good. When we talk about hallucinations, or weird outputs, or models doing unexpected and unwanted things, certainly…
You still can’t measure brand strength in AI. Ashley Faus pointed out this morning on another post that a company is claiming to measure your brand strength/presence in AI. Unsurprisingly, the claim was not accompanied with a peer-reviewed dataset or methodology, or even a preprint paper. Let’s review why this is impossible today. Models are…
Summary In today's episode, I explain how to cultivate connections with AI by building specialized knowledge blocks before you begin a conversation. Here's what this means for you. You transform a naive AI assistant into a well-informed thought partner that can actively challenge your perspective. You'll also learn these concepts: how personal and professional data…
Summary In today's episode, I explain how to use priming to transform a naive AI into an expert thinking partner. Here's what this means for you. You gain the ability to drive much more intelligent and specialized discussions by providing high-quality context. You'll also learn these concepts: how to build information-dense knowledge blocks, the value…
Who will benefit from DeepSeek Open Week? DeepSeek, the Chinese AI company, has been giving away code this week on their Github account. So far, they’ve released: FlashMLA: Efficient MLA Decoding Kernel DeepEP: the first open-source EP communication library for MoE model training and inference. Expert Parallelism Load Balancer: efficient load balancing for experts in…
