Expressing Intent with LLMs

Today we talk to LLMs like to a child, in simple low-context text; as models grow up, they should grasp more intent with fewer words.

Contents
    Illustration of three small childlike figures with backpacks climbing a set of round steps toward a glowing, flower-like orb, beneath a giant detailed brain floating in a digital landscape.

    Use it when Deciding how much to invest in prompt engineering versus work that goes beyond the prompt.

    Now: talking to a child. Chatting with an LLM today means breaking what you want into simple, concise pieces of text the model can relate to its training data. This is low-context communication. We compensate with prompt engineering, manual reviews of inputs and sentence re-ordering, and each model needs to be prompted differently depending on how it was trained.

    Humans are different. Adults understand more intent with fewer words (high-context communication). To us, prompts are not just text but also smell, vision and other things we experienced through our senses; they are how we access memory. LLMs rely on text alone to trigger their memory.

    Future: models grow up. They may need little or no prompt engineering, understand language and culture, learn each person’s nuances, and even have senses (today there is vision).

    The open question: do we build knowledge stacks around prompt engineering and make an art of it, or assume models will get smarter and focus on what lies beyond prompts?

    Topics

    Agents in Production 10Building with AI 9AI Product Strategy 24

    Start here

    How Agents Run in ProductionBeyond LLMs: The Whole Product Framework for AIThe Verification Bottleneck: Why AI's Real Cost Is Human Attention

    ↑ ↓ to move, Enter for all results