Tag: LLM

  • An emerging discipline?

    Transparency label: AI-assisted

    In a recent post, I observed that LLMs are coming to be seen as like computer operating systems, with prompts as the new application programs and context as the new user interface.

    Support for that way of thinking comes from how we can use Anapoly’s precision content prompt pack to apply structure to a source document. The pack contains a set of prompts which are applied in sequence, with human validation of the AI’s output at each step in the sequence. In effect, it is a program governing the end-to-end interaction between AI and human. 

    Contract-first prompting takes the idea further. It formalises the interaction between human and AI into a negotiated agreement, locking in intent, scope, constraints, and deliverables before any output is generated. This structured handshake resembles a statement of work in engineering or a project brief in consulting. It ensures both sides – human and AI – share the same understanding of the task. Compliance mechanisms to be used during the conduct of work – such as summarisation, clarifying loops, and self-testing – are also built into the agreement. Thus, the contract becomes both compass and checklist.

    Our precision content prompt pack and the contract-first idea transform prompting into programmable context design.

    This reframes the interaction between human and AI: not as issuing commands, but as engineering the conditions under which the AI will perform. When we treat prompts, inputs, roles, and structure as modular components of a working system, we begin to move from improvisation toward disciplined practice. Not just better prompting – but contextual systems engineering.

    Transparency label justification: This diary post was drafted by Alec, with ChatGPT used to suggest edits, refine wording, and test the logic of specific formulations. Alec initiated the framing, decided the sequence of ideas, and approved the final structure and terminology.

  • Context is the new user interface

    Transparency label: Human-led

    How we guide AI not with buttons, but with words and context.

    When you use an AI like ChatGPT, there are no menus or toolbars. Instead, the way you interact – what you say, how you say it, and what you show it – is the interface.

    That’s what we mean by context. It’s the information you give the AI to help it understand what you want. A vague prompt gives you a vague answer. A clear prompt, with the right background, gets you something much more useful.

    This new kind of interface has several parts:

    • Prompt window – where you ask questions and give tasks
    • Custom instructions – where you shape the AI’s behaviour
    • File uploads – where you provide background material
    • Project spaces – where you store persistent context for a task
    • Canvas – where you co-develop ideas with the AI

    You don’t click buttons. You build context. And the better the context, the better the outcome.


    Backlink: An LLM is like an operation system

    This post was written by Alec with support from ChatGPT. Alec directed the concept, structure, and final wording. ChatGPT contributed phrasings and refinements, which were reviewed and approved by Alec.

  • An LLM is like an operating system

    Transparency label: human-led

    Reference: Andrej Karpathy’s keynote address on 17 June 2025 at AI Startup School in San Francisco

    In the 1960s and ’70s:

    • a computer filled a large room in a central location, far from most of its users
    • it had an operating system which provided basic functions
    • application programs used operating system functions to do useful things
    • we accessed the computer remotely over telephone lines
    • we communicated with the computer using only text, from a terminal

    In 2010:

    • computers were personal, portable, and mostly ran local software
    • the operating system managed files, programs, and user settings
    • applications were installed and launched directly by the user
    • internet access became always-on, but most processing still happened locally
    • we communicated with the computer through keyboard, mouse, and graphical interface

    In 2025:

    • an AI runs in a cloud-based data centre, far from the user
    • it has an LLM which provides core functions for language, reasoning, and knowledge work
    • prompts use the core functions to do useful things
    • we access the AI remotely over the internet
    • we communicate with it using mainly text, (voice, or images are also possible, now) and still mostly from a terminal.

    In 2050:

    • an AI will be personal, portable, and …

    LLMs are the new operating systems.

    Prompts are the new application programs.

    Context is the new user interface.

    Transparency: This post was conceived, structured, and written by a human author. ChatGPT was used to suggest phrasing and refine analogies, but all key ideas, narrative framing, and editorial decisions were made by the human. AI input was limited to supporting the writing process.