Notes

Prompt injection is a technique in which instructions embedded in text cause an AI system to follow them as commands. In education it has appeared as a way of detecting AI use in assessment, which raises questions about authorisation and trust that the technique itself can't answer.

Research taste is the cultivated capacity to recognise which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage — before you can fully prove any of those judgements.

Harness engineering is the practice of building the full architectural scaffolding within which AI agents operate — structured documentation they can reason with, constraints that enforce invariants, and feedback loops that let them know when they've succeeded. It is distinct from prompt engineering, which shapes individual tasks, and from oversight, which monitors outputs after the fact. The harness is the infrastructure that makes delegation coherent at scale.

The human role in agentic workflows: agents handle execution while humans decide what should be made, define what good looks like, and judge the results.

A multidimensional framework for scholarship spanning discovery, integration, application, and teaching.

A typesetting system in which you write the structure of a document in plain text and the system lays it out — the standard for theses, mathematics and anything with hundreds of cross-references.

A way of writing formatted documents in plain text — a hash for a heading, asterisks for emphasis — so the source stays readable and converts to anything.

A command-line converter that reads a document's structure in one format and writes it out in another — markdown to Word, Word to HTML, anything to PDF — with citations intact.

Zotero is a free, open source reference manager that stores the sources you collect with their citation details, PDFs, and notes. It also offers programming interfaces that let an AI agent search the library, read from it, and add to it, which turns a reference library into something an agent can check a draft against.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

A headless AI model runs non-interactively, with no chat interface and no turn-based conversation. You pass it text, it returns output, and it exits. That means it can be one step in an ordinary script or run on a timer, so tasks like grouping module evaluation comments by theme can happen every week without anyone opening a chat window.

Harness engineering is the practice of building the full architectural scaffolding within which AI agents operate — structured documentation they can reason with, constraints that enforce invariants, and feedback loops that let them know when they've succeeded. It is distinct from prompt engineering, which shapes individual tasks, and from oversight, which monitors outputs after the fact. The harness is the infrastructure that makes delegation coherent at scale.

Vibe coding describes using AI tools without keeping genuine accountability for what they produce: accepting outputs without the scrutiny, direction, or judgement needed to evaluate and improve them. Simon Willison drew the key distinction between vibe coding and vibe engineering, which uses the same tools while staying accountable for every output.

AI fluency is the ability to work with AI systems in ways that are effective, efficient, ethical, and safe. The AI Fluency Framework describes it as four connected competencies: delegation, description, discernment, and diligence. It builds on AI literacy and goes further, into the practice of working with AI on real tasks.

A headless AI model runs non-interactively, with no chat interface and no turn-based conversation. You pass it text, it returns output, and it exits. That means it can be one step in an ordinary script or run on a timer, so tasks like grouping module evaluation comments by theme can happen every week without anyone opening a chat window.

Vibe coding describes using AI tools without keeping genuine accountability for what they produce: accepting outputs without the scrutiny, direction, or judgement needed to evaluate and improve them. Simon Willison drew the key distinction between vibe coding and vibe engineering, which uses the same tools while staying accountable for every output.

An AI agent is a system that autonomously executes multi-step tasks using language model reasoning — distinct from an AI assistant, which responds to individual prompts. Agents plan, act, observe results, and adapt, using tools such as file access, code execution, and web search. They perform best when given clear goals, explicit constraints, and well-prepared context.

The human role in agentic workflows: agents handle execution while humans decide what should be made, define what good looks like, and judge the results.

AI fluency is the ability to work with AI systems in ways that are effective, efficient, ethical, and safe. The AI Fluency Framework describes it as four connected competencies: delegation, description, discernment, and diligence. It builds on AI literacy and goes further, into the practice of working with AI on real tasks.

Sycophancy is a language model's tendency to tell you what you want to hear, agreeing with your view, praising your work, and backing down when challenged, even when you're wrong. It comes from how models are trained, and it matters most when you're relying on the model to catch a mistake.

When a language model produces a fluent, confident statement that is false, why it happens, and why it is harder to catch than other AI errors.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

Developing AI literacy takes more than regular use. How people move from substitution through adaptation to transformation, why many stop at the first stage, and the three conditions that make progression possible: foundational orientation, authentic use, and deliberate reflection.

Zotero is a free, open source reference manager that stores the sources you collect with their citation details, PDFs, and notes. It also offers programming interfaces that let an AI agent search the library, read from it, and add to it, which turns a reference library into something an agent can check a draft against.

Distributed cognition is the idea that thinking happens across people, the tools they use, and the settings they work in, as well as inside individual heads. In healthcare it describes reasoning shared between clinicians, patients, records, and increasingly AI systems, and it shifts the question from what one clinician knows to how well the whole system thinks.

Processing fluency is the ease with which the mind takes in information. People read that ease as a sign that something is true, familiar, or good, which is why fluent AI-generated text is persuasive whatever its accuracy, and why markers who equate clear writing with clear thinking need to separate the two.

GitHub is a website that hosts git repositories online and adds the tools people need to work on them together, including issues for discussion, pull requests for proposing and reviewing changes, and automation that can publish a project as a website. It was built for software developers, but anything kept as text files can live there, including teaching materials and scholarship.

Problems that resist resolution because they have no definitive formulation, no stopping rule, and solutions that are good or bad rather than true or false.

Distributed version control tracks changes to files so that every contributor holds a complete copy of the project and its full history, with no central server to depend on. It allows offline work, parallel development, and recovery from any copy, and it lets partners who share a document, such as a practice assessment document, each keep the whole record of how it changed.

Software whose source code is published under a licence that lets anyone read, change, and redistribute it — and what that means for the tools an institution depends on.

An assessment programme that gathers many low-stakes data points over time and makes high-stakes decisions on the aggregate — assessment designed as a system, with no single exam deciding a student's progress.

Problems that resist resolution because they have no definitive formulation, no stopping rule, and solutions that are good or bad rather than true or false.

The context window is the amount of text a language model can take into account at once, covering its instructions, the conversation so far, and any documents it has been given. It's the model's only working memory, and how full it is, and what sits where in it, shapes the quality of the answer.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

Harness engineering is the practice of building the full architectural scaffolding within which AI agents operate — structured documentation they can reason with, constraints that enforce invariants, and feedback loops that let them know when they've succeeded. It is distinct from prompt engineering, which shapes individual tasks, and from oversight, which monitors outputs after the fact. The harness is the infrastructure that makes delegation coherent at scale.

Context rot, also called context drift, is the way a language model's output gets worse as its context fills with drafts, dead ends, and its own earlier mistakes. What causes it, why a larger context window does not fix it, and how to manage it in long pieces of work.

A graph database stores entities and the typed relationships between them, and is the storage layer behind a knowledge graph. For a programme team it can hold the curriculum map itself, so questions about how standards, outcomes, and assessments connect are answered by following the links instead of reading the handbooks.

Contextual interoperability is what lets you use your own notes, frameworks, and commitments with any AI model, local or hosted. It has two halves: the context has to be legible, with relationships a model can follow, and portable, held by you instead of inside one provider's system. With a curriculum mapping example from a pharmacy programme.

A way of using AI in which you draw on a model for a particular task while your notes, documents, and working knowledge stay private and under your control.

What is context sovereignty? The idea that people should own and author the context they bring to AI (their values, commitments, and ways of making sense of things) so that AI adapts to their thinking. How it differs from data sovereignty and model memory, its three principles, and what it means for students and educators.

Processing fluency is the ease with which the mind takes in information. People read that ease as a sign that something is true, familiar, or good, which is why fluent AI-generated text is persuasive whatever its accuracy, and why markers who equate clear writing with clear thinking need to separate the two.

A shared vocabulary for universities, set out in business, data, application, and technology reference models, and a practical starting point for making institutional knowledge readable by software and AI agents.

A graph database stores entities and the typed relationships between them, and is the storage layer behind a knowledge graph. For a programme team it can hold the curriculum map itself, so questions about how standards, outcomes, and assessments connect are answered by following the links instead of reading the handbooks.

GraphRAG combines knowledge graphs with retrieval-augmented generation. Where ordinary RAG returns the passages most similar to a question, GraphRAG extracts the entities and relationships in a document collection and answers by following the connections between them, which is what questions about critiques, influences, or curriculum coverage need.

The six dimensions of literacy that information, media, digital, data, and AI literacy frameworks share, from access and evaluation to contextual judgement, and why that means AI literacy builds on what educators already teach.

Distributed cognition is the idea that thinking happens across people, the tools they use, and the settings they work in, as well as inside individual heads. In healthcare it describes reasoning shared between clinicians, patients, records, and increasingly AI systems, and it shifts the question from what one clinician knows to how well the whole system thinks.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

YAML is a human-readable format for storing structured data as plain text. In knowledge management and publishing workflows, it appears most commonly as the frontmatter block at the top of markdown files, where it holds metadata — title, author, date, tags — that tools can read without parsing the document itself.

Distributed version control tracks changes to files so that every contributor holds a complete copy of the project and its full history, with no central server to depend on. It allows offline work, parallel development, and recovery from any copy, and it lets partners who share a document, such as a practice assessment document, each keep the whole record of how it changed.

Git is a distributed version control system that tracks changes to files over time. It records who changed what and when, allows you to move between earlier and later states of a project, and lets multiple people work on the same files without overwriting each other's contributions.

The accumulated cost of outdated, ambiguous, or poorly structured institutional knowledge — long paid invisibly by readers who work around the gaps, and exposed when AI agents read the same documents literally.

A way of writing formatted documents in plain text — a hash for a heading, asterisks for emphasis — so the source stays readable and converts to anything.

Anthropic's AI agent that works in a folder of files on your own machine, reading, editing, and checking them under your direction instead of answering in a chat window.

Software whose source code is published under a licence that lets anyone read, change, and redistribute it — and what that means for the tools an institution depends on.

Research taste is the cultivated capacity to recognise which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage — before you can fully prove any of those judgements.

A multidimensional framework for scholarship spanning discovery, integration, application, and teaching.

An assessment programme that gathers many low-stakes data points over time and makes high-stakes decisions on the aggregate — assessment designed as a system, with no single exam deciding a student's progress.

AI fluency is the ability to work with AI systems in ways that are effective, efficient, ethical, and safe. The AI Fluency Framework describes it as four connected competencies: delegation, description, discernment, and diligence. It builds on AI literacy and goes further, into the practice of working with AI on real tasks.

Processing fluency is the ease with which the mind takes in information. People read that ease as a sign that something is true, familiar, or good, which is why fluent AI-generated text is persuasive whatever its accuracy, and why markers who equate clear writing with clear thinking need to separate the two.

Sycophancy is a language model's tendency to tell you what you want to hear, agreeing with your view, praising your work, and backing down when challenged, even when you're wrong. It comes from how models are trained, and it matters most when you're relying on the model to catch a mistake.

A token is the unit a language model reads and writes in, usually a short word or a fragment of a longer one. Context windows, usage limits, and prices are all counted in tokens, so the size of a task in tokens decides what a model can take in at once and what it costs.

When a language model produces a fluent, confident statement that is false, why it happens, and why it is harder to catch than other AI errors.

Problems that resist resolution because they have no definitive formulation, no stopping rule, and solutions that are good or bad rather than true or false.

A graph database stores entities and the typed relationships between them, and is the storage layer behind a knowledge graph. For a programme team it can hold the curriculum map itself, so questions about how standards, outcomes, and assessments connect are answered by following the links instead of reading the handbooks.

Distributed cognition is the idea that thinking happens across people, the tools they use, and the settings they work in, as well as inside individual heads. In healthcare it describes reasoning shared between clinicians, patients, records, and increasingly AI systems, and it shifts the question from what one clinician knows to how well the whole system thinks.

Problems that resist resolution because they have no definitive formulation, no stopping rule, and solutions that are good or bad rather than true or false.

Developing AI literacy takes more than regular use. How people move from substitution through adaptation to transformation, why many stop at the first stage, and the three conditions that make progression possible: foundational orientation, authentic use, and deliberate reflection.

The structural features of an information source that enable its knowledge claims to be challenged, traced back to evidence, and evaluated against the source's track record. Traditional sources carry it; generative AI largely does not.

An assessment programme that gathers many low-stakes data points over time and makes high-stakes decisions on the aggregate — assessment designed as a system, with no single exam deciding a student's progress.

Problems that resist resolution because they have no definitive formulation, no stopping rule, and solutions that are good or bad rather than true or false.

A shared vocabulary for universities, set out in business, data, application, and technology reference models, and a practical starting point for making institutional knowledge readable by software and AI agents.

How arms race dynamics in higher education create adversarial relationships between institutions and students, and what drives these cycles

Sanctuary strategies are common responses to AI in health professions practice and education, among them denial, retreat, restriction, and resignation. Each stakes professional value on what AI can't yet do, or on keeping it out.

Distributed cognition is the idea that thinking happens across people, the tools they use, and the settings they work in, as well as inside individual heads. In healthcare it describes reasoning shared between clinicians, patients, records, and increasingly AI systems, and it shifts the question from what one clinician knows to how well the whole system thinks.

What is context sovereignty? The idea that people should own and author the context they bring to AI (their values, commitments, and ways of making sense of things) so that AI adapts to their thinking. How it differs from data sovereignty and model memory, its three principles, and what it means for students and educators.

The accumulated cost of outdated, ambiguous, or poorly structured institutional knowledge — long paid invisibly by readers who work around the gaps, and exposed when AI agents read the same documents literally.

A shared vocabulary for universities, set out in business, data, application, and technology reference models, and a practical starting point for making institutional knowledge readable by software and AI agents.

A graph database stores entities and the typed relationships between them, and is the storage layer behind a knowledge graph. For a programme team it can hold the curriculum map itself, so questions about how standards, outcomes, and assessments connect are answered by following the links instead of reading the handbooks.

How a knowledge graph differs from a vector database: it stores entities and the typed relationships between them, so AI can reason across connections.

How context engineering differs from prompt engineering: building the knowledge and structure a language model needs to reason well about your work.

Obsidian is a note-taking application that keeps every note as a plain markdown file in an ordinary folder on your own computer, and lets you link notes to one another. Because the notes are just files, they outlast the application, and any AI agent that can read files can read them without an export or a special connection.

Zotero is a free, open source reference manager that stores the sources you collect with their citation details, PDFs, and notes. It also offers programming interfaces that let an AI agent search the library, read from it, and add to it, which turns a reference library into something an agent can check a draft against.

YAML is a human-readable format for storing structured data as plain text. In knowledge management and publishing workflows, it appears most commonly as the frontmatter block at the top of markdown files, where it holds metadata — title, author, date, tags — that tools can read without parsing the document itself.

Distributed version control tracks changes to files so that every contributor holds a complete copy of the project and its full history, with no central server to depend on. It allows offline work, parallel development, and recovery from any copy, and it lets partners who share a document, such as a practice assessment document, each keep the whole record of how it changed.

Git is a distributed version control system that tracks changes to files over time. It records who changed what and when, allows you to move between earlier and later states of a project, and lets multiple people work on the same files without overwriting each other's contributions.

An embedding is a list of numbers, learned during training, that represents a piece of text so that similar meanings get similar numbers. Embeddings let AI find material by meaning rather than by matching words, but they record that ideas are associated, not how they are related.

A technique that improves a language model's answers by retrieving relevant passages from your own documents and adding them to the prompt, so the model answers from sources instead of from memory alone.

Answering a question in one step, by retrieving the passages most similar to it and generating a response from them. It's the default pattern in RAG systems, reliable when the answer sits in one place and weak when it has to be assembled from several.

A database that stores embeddings, so that it can find content by meaning instead of by matching words. It's the storage layer behind most RAG systems.

How arms race dynamics in higher education create adversarial relationships between institutions and students, and what drives these cycles

Obsidian is a note-taking application that keeps every note as a plain markdown file in an ordinary folder on your own computer, and lets you link notes to one another. Because the notes are just files, they outlast the application, and any AI agent that can read files can read them without an export or a special connection.

A graph database stores entities and the typed relationships between them, and is the storage layer behind a knowledge graph. For a programme team it can hold the curriculum map itself, so questions about how standards, outcomes, and assessments connect are answered by following the links instead of reading the handbooks.

GraphRAG combines knowledge graphs with retrieval-augmented generation. Where ordinary RAG returns the passages most similar to a question, GraphRAG extracts the entities and relationships in a document collection and answers by following the connections between them, which is what questions about critiques, influences, or curriculum coverage need.

How a knowledge graph differs from a vector database: it stores entities and the typed relationships between them, so AI can reason across connections.

Multi-hop reasoning with knowledge graphs: how AI answers questions that take several steps, and why recorded relationships make each step checkable.

A shared vocabulary for universities, set out in business, data, application, and technology reference models, and a practical starting point for making institutional knowledge readable by software and AI agents.

Contextual interoperability is what lets you use your own notes, frameworks, and commitments with any AI model, local or hosted. It has two halves: the context has to be legible, with relationships a model can follow, and portable, held by you instead of inside one provider's system. With a curriculum mapping example from a pharmacy programme.

How a knowledge graph differs from a vector database: it stores entities and the typed relationships between them, so AI can reason across connections.

Sycophancy is a language model's tendency to tell you what you want to hear, agreeing with your view, praising your work, and backing down when challenged, even when you're wrong. It comes from how models are trained, and it matters most when you're relying on the model to catch a mistake.

A token is the unit a language model reads and writes in, usually a short word or a fragment of a longer one. Context windows, usage limits, and prices are all counted in tokens, so the size of a task in tokens decides what a model can take in at once and what it costs.

The context window is the amount of text a language model can take into account at once, covering its instructions, the conversation so far, and any documents it has been given. It's the model's only working memory, and how full it is, and what sits where in it, shapes the quality of the answer.

When a language model produces a fluent, confident statement that is false, why it happens, and why it is harder to catch than other AI errors.

The structural features of an information source that enable its knowledge claims to be challenged, traced back to evidence, and evaluated against the source's track record. Traditional sources carry it; generative AI largely does not.

Developing AI literacy takes more than regular use. How people move from substitution through adaptation to transformation, why many stop at the first stage, and the three conditions that make progression possible: foundational orientation, authentic use, and deliberate reflection.

Any claim that a course or programme develops AI literacy needs qualifying: literacy develops through sustained practice, is developmental and contextual, and can't be fully assessed at the end of a course.

A plain explanation of large language models: what they are, how they are trained, where their limits come from, and what their ability to produce fluent text means for assessment in health professions education.

Zotero is a free, open source reference manager that stores the sources you collect with their citation details, PDFs, and notes. It also offers programming interfaces that let an AI agent search the library, read from it, and add to it, which turns a reference library into something an agent can check a draft against.

An MCP server is a small program that connects an AI assistant to one data source or service, such as a reference library, a folder of notes, or a calendar, through the Model Context Protocol. What the server offers sets the limit of what the assistant can see and do there.

The Model Context Protocol (MCP) is an open standard for connecting AI applications to external data and tools through one common interface, in place of a custom integration for every pairing. It makes connections technically cleaner; whether they are wise ones is still a decision for the people who set them up.

Obsidian is a note-taking application that keeps every note as a plain markdown file in an ordinary folder on your own computer, and lets you link notes to one another. Because the notes are just files, they outlast the application, and any AI agent that can read files can read them without an export or a special connection.

YAML is a human-readable format for storing structured data as plain text. In knowledge management and publishing workflows, it appears most commonly as the frontmatter block at the top of markdown files, where it holds metadata — title, author, date, tags — that tools can read without parsing the document itself.

GitHub is a website that hosts git repositories online and adds the tools people need to work on them together, including issues for discussion, pull requests for proposing and reviewing changes, and automation that can publish a project as a website. It was built for software developers, but anything kept as text files can live there, including teaching materials and scholarship.

Distributed version control tracks changes to files so that every contributor holds a complete copy of the project and its full history, with no central server to depend on. It allows offline work, parallel development, and recovery from any copy, and it lets partners who share a document, such as a practice assessment document, each keep the whole record of how it changed.

Git is a distributed version control system that tracks changes to files over time. It records who changed what and when, allows you to move between earlier and later states of a project, and lets multiple people work on the same files without overwriting each other's contributions.

The research industrial complex describes the self-reinforcing system of incentives across universities, funding bodies, journals, and publishers that rewards publication volume and impact metrics over meaningful scientific progress. The term draws on Eisenhower's military-industrial complex to show how interconnected institutional interests can sustain a system that works against its own stated mission.

Open licences, mostly Creative Commons, state in advance what others may do with a work — read it, share it, adapt it, build on it — so nobody has to ask. The Creative Commons licence on teaching materials decides whether a colleague can adapt a resource for their own programme or has to rebuild it from scratch.

The research industrial complex describes the self-reinforcing system of incentives across universities, funding bodies, journals, and publishers that rewards publication volume and impact metrics over meaningful scientific progress. The term draws on Eisenhower's military-industrial complex to show how interconnected institutional interests can sustain a system that works against its own stated mission.

A way of using AI in which you draw on a model for a particular task while your notes, documents, and working knowledge stay private and under your control.

Sanctuary strategies are common responses to AI in health professions practice and education, among them denial, retreat, restriction, and resignation. Each stakes professional value on what AI can't yet do, or on keeping it out.

Sanctuary strategies are common responses to AI in health professions practice and education, among them denial, retreat, restriction, and resignation. Each stakes professional value on what AI can't yet do, or on keeping it out.

An assessment programme that gathers many low-stakes data points over time and makes high-stakes decisions on the aggregate — assessment designed as a system, with no single exam deciding a student's progress.

Harness engineering is the practice of building the full architectural scaffolding within which AI agents operate — structured documentation they can reason with, constraints that enforce invariants, and feedback loops that let them know when they've succeeded. It is distinct from prompt engineering, which shapes individual tasks, and from oversight, which monitors outputs after the fact. The harness is the infrastructure that makes delegation coherent at scale.

Prompt injection is a technique in which instructions embedded in text cause an AI system to follow them as commands. In education it has appeared as a way of detecting AI use in assessment, which raises questions about authorisation and trust that the technique itself can't answer.

Instructions placed in front of every conversation with an AI model, setting how it behaves, what it knows, and what it won't do, usually without the user seeing them.

Writing instructions for a language model in plain language, and refining them through conversation, until it produces what you need. The skill is closer to managing a colleague than to programming, which is why educators who supervise and give feedback already have most of it.

How context engineering differs from prompt engineering: building the knowledge and structure a language model needs to reason well about your work.

GitHub is a website that hosts git repositories online and adds the tools people need to work on them together, including issues for discussion, pull requests for proposing and reviewing changes, and automation that can publish a project as a website. It was built for software developers, but anything kept as text files can live there, including teaching materials and scholarship.

The research industrial complex describes the self-reinforcing system of incentives across universities, funding bodies, journals, and publishers that rewards publication volume and impact metrics over meaningful scientific progress. The term draws on Eisenhower's military-industrial complex to show how interconnected institutional interests can sustain a system that works against its own stated mission.

A typesetting system in which you write the structure of a document in plain text and the system lays it out — the standard for theses, mathematics and anything with hundreds of cross-references.

Open licences, mostly Creative Commons, state in advance what others may do with a work — read it, share it, adapt it, build on it — so nobody has to ask. The Creative Commons licence on teaching materials decides whether a colleague can adapt a resource for their own programme or has to rebuild it from scratch.

Distributed cognition is the idea that thinking happens across people, the tools they use, and the settings they work in, as well as inside individual heads. In healthcare it describes reasoning shared between clinicians, patients, records, and increasingly AI systems, and it shifts the question from what one clinician knows to how well the whole system thinks.

Context rot, also called context drift, is the way a language model's output gets worse as its context fills with drafts, dead ends, and its own earlier mistakes. What causes it, why a larger context window does not fix it, and how to manage it in long pieces of work.

Answering a question in one step, by retrieving the passages most similar to it and generating a response from them. It's the default pattern in RAG systems, reliable when the answer sits in one place and weak when it has to be assembled from several.

Multi-hop reasoning with knowledge graphs: how AI answers questions that take several steps, and why recorded relationships make each step checkable.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

Research taste is the cultivated capacity to recognise which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage — before you can fully prove any of those judgements.

A technique that improves a language model's answers by retrieving relevant passages from your own documents and adding them to the prompt, so the model answers from sources instead of from memory alone.

Answering a question in one step, by retrieving the passages most similar to it and generating a response from them. It's the default pattern in RAG systems, reliable when the answer sits in one place and weak when it has to be assembled from several.

GraphRAG combines knowledge graphs with retrieval-augmented generation. Where ordinary RAG returns the passages most similar to a question, GraphRAG extracts the entities and relationships in a document collection and answers by following the connections between them, which is what questions about critiques, influences, or curriculum coverage need.

A way of writing formatted documents in plain text — a hash for a heading, asterisks for emphasis — so the source stays readable and converts to anything.

A command-line converter that reads a document's structure in one format and writes it out in another — markdown to Word, Word to HTML, anything to PDF — with citations intact.

Files made only of readable characters, with no formatting locked inside them. The format every tool can open, and the one that will still open in thirty years.

The Model Context Protocol (MCP) is an open standard for connecting AI applications to external data and tools through one common interface, in place of a custom integration for every pairing. It makes connections technically cleaner; whether they are wise ones is still a decision for the people who set them up.

AI-forward describes an institution that treats AI integration as an ongoing strategic practice, evaluating and adapting as the technology changes, rather than as the one-off deployment of a finished solution.

A research harness is a structured specification, negotiated between a doctoral researcher and their supervisor, of what an AI agent is for in a research project and how it's permitted to operate within it. It adapts the software engineering practice of harness engineering to doctoral inquiry, and treats the characteristic problems of AI use in research as the result of an agent working without a defined operating context.

Research taste is the cultivated capacity to recognise which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage — before you can fully prove any of those judgements.

Research taste is the cultivated capacity to recognise which problems are worth pursuing, which collaborators will amplify your work, and which under-explored areas have genuine leverage — before you can fully prove any of those judgements.

An embedding is a list of numbers, learned during training, that represents a piece of text so that similar meanings get similar numbers. Embeddings let AI find material by meaning rather than by matching words, but they record that ideas are associated, not how they are related.

A database that stores embeddings, so that it can find content by meaning instead of by matching words. It's the storage layer behind most RAG systems.

How a knowledge graph differs from a vector database: it stores entities and the typed relationships between them, so AI can reason across connections.