How AI works

What large language models are, what they can and cannot do, and how the field has changed since ChatGPT.

2026 13

Four Years After ChatGPT

Talks University of Oxford AI Competency Centre · 30 November 2026 · Webinar

Dominik's annual retrospective webinar for the University of Oxford AI Competency Centre.

What's new in AI this month?

Talks MondAI Roundup · 16 November 2026 and 14 December 2026 · Webinar

A monthly roundup of AI news, with an RSS feed.

The future of data in the age of AI Agents: New opportunities for old methods in data management and collection

Talks Oxford Data Meetup · 12 November 2026 · Talk

We are at an inflection point in what transformation we can expect in the age of AI. Agentic AI systems are now capable and reliable in way that allows us to delegate meaningful work to them. In the second half of 2026 we now have…

AI Thought Exchange, October meeting

Talks AI Thought Exchange · 20 October 2026 · Talk

Dominik opens the October meeting of the AI Thought Exchange with a short review of what happened in AI over the summer, followed by discussion with educators from the UK and Australia. His slot starts at 09:10 BST. The AI Thought…

Common misconceptions about Large Language Models and a few literal metaphors

Writing Schemas & Propositions · 4 August 2026 · Blog post

The explainer corrects common misconceptions about language-model architecture, memory, capabilities, training, tool use, and emotional representation.

York Festival of AI: nine sessions

Talks York Festival of AI · 20–22 July 2026 · Talk

Teaching During the Cognitive Revolution

Talks Education in an AI World, Jersey College for Girls · 26 June 2026 · Talk

The emergence of Large Language Models (LLMs) has completely upset not only what we can expect of a computer but also how we view the role of technology in the process of learning. Now we have machines that can help at all stages of the…

AI and Expertise

Talks FEDA Education Gala · 14 June 2026 · Talk

Language without structure? How can LLMs speak all the languages without knowing any of the rules?

Talks Wolfson College, OxfordXML seminar "Human Language Meets AI" · 30 April 2026 · Talk

Panel at the OECD

Talks OECD · 6 March 2026 · Talk

Seminar on AI and programming at the Mathematical Institute

Talks Mathematical Institute · 20 February 2026 · Talk

We've been using AI wrong: Agent Skills and the Semantic Powers of LLMs

Writing Schemas & Propositions · 2 February 2026 · Blog post

The guide explains how Agent Skills combine language-model judgement with reusable instructions, files, and scripts to support reliable workflows within context and memory limits.

Teaching during the Cognitive Revolution: What will learning and teaching look like in the age of AI?

Talks MPLS Researcher Conference · 15 January 2026 · Talk

The emergence of Large Language Models has disrupted expectations of computing and technology's role in learning. This session explores how AI can support learners and educators at every stage, from managing cognitive load to serving as…

2025 20

How Humans Hallucinate and How We Deal with It

Writing Schemas & Propositions · 7 December 2025 · Blog post

The essay treats ordinary human errors as hallucinations and explains how metacognition, tools, institutions, and social norms constrain them.

Three Years After ChatGPT: From chatbots to agents and other trends on the frontier of generative AI

Talks Microsoft Teams · 2 December 2025 · Webinar

Three years after the public launch of ChatGPT, the chatbot paradigm has undergone a transformation. The main mode of interacting with Large Language Models is no longer just a chat. This session steps back from day-to-day trends to…

Three Years After ChatGPT: From Chatbots to Agents and the Shifting Frontier of Generative AI

Writing AI in Academic Practice (LinkedIn) · 27 November 2025 · Blog post

How LLMs Understand Text: Context and Meaning in the AI Age

Talks Kellogg College, Oxford · 17 November 2025 · Talk

Large Language Models represent an entirely new paradigm of semantic information processing. They are the first time a computer can process the meaning of text without mark-up. Their ability to 'understand' text is at the same time…

Symbols and Manipulation

Writing Schemas & Propositions · 8 September 2025 · Blog post

The essay argues that human and language-model cognition operates through meaning-rich schemas rather than the formal symbol manipulation performed by computers.

Karpathy Thinks in Code

Writing Schemas & Propositions · 31 August 2025 · Blog post

The essay argues that humans and language models interpret code through schemas while computers execute formal instructions without semantic understanding.

50 First Dates and the ChatGPT Memory Problem

Writing Schemas & Propositions · 27 August 2025 · Blog post

The essay argues that language-model memory consists of propositional notes that cannot reproduce the tacit schemas formed through experience.

GPT-5 is here! Long time coming but blazingly fast!

Writing AI in Academic Practice (LinkedIn) · 9 August 2025 · Blog post

What is ChatGPT Agent and what does it tell us about the future of AI?

Writing AI in Academic Practice (LinkedIn) · 18 July 2025 · Blog post

OPP Summer Lecture

Talks July 2025 · Talk

Embarrassingly out of date and unreliable! Some thoughts on the new Stanford AI Index Report

Writing AI in Academic Practice (LinkedIn) · 13 April 2025 · Blog post

Creation is perception: On the role of skill in the age of AI

Writing AI in Academic Practice (LinkedIn) · 12 April 2025 · Blog post

AI Fluency as a Job Requirement: Karpathy on Individual Advantage and Shopify's New Mandate

Writing AI in Academic Practice (LinkedIn) · 8 April 2025 · Blog post

All art is conceptual art: Reflections on the aftershocks of an AI image-generation earthquake

Writing AI in Academic Practice (LinkedIn) · 1 April 2025 · Blog post

"You're doing AI wrong": Choosing an AI tool is more like hiring an employee than buying software

Writing AI in Academic Practice (LinkedIn) · 30 March 2025 · Blog post

GPT-4o's native image generation is a true revolution in what is possible with AI and images

Writing AI in Academic Practice (LinkedIn) · 25 March 2025 · Blog post

Talk on AI at OxLearn

Talks OxLearn · 25 February 2025 · Talk

The future of tech is human: Redefining the role of educators in the wake of Industry 5.0

Talks Bett 2025, Innovation Theatre · 24 January 2025 · Talk

Two years after ChatGPT

Talks Bett 2025, ExCeL London · 24 January 2025 · Talk

What will 2025 be like for AI? 5 Predictions for the Future

Writing AI in Academic Practice (LinkedIn) · 1 January 2025 · Blog post

2024 10

A Year in AI: Doing more with less in 2024

Writing AI in Academic Practice (LinkedIn) · 31 December 2024 · Blog post

Two Years After ChatGPT: Reflections on The State and Future Directions of Generative AI

Talks Microsoft Teams · 28 November 2024 · Webinar

On the two-year anniversary of the release of ChatGPT, this webinar reflects on the journey to today and the current state of generative AI with a focus on what has changed compared to our expectations in 2022, what has stayed the same…

What have Large Language Models actually learned about language

Talks Wolfson College, OxfordXML · 26 November 2024 · Talk

This talk will explore what exactly large language models (LLMs) have learned about languages, and why it matters. It will attempt an outline of an answer to how these models understand the structure of language, and how this compares…

Magical Meaning Machines

Talks Centre for Digital Scholarship, Weston Library · 20 November 2024 · Talk

This practical workshop will explore the opportunities for new kinds of digital scholarship in the humanities and social sciences presented by Large Language Models. It will present Large Language Models as a new kind of machine for…

Trend 2: Interface Innovations: (Re)defining the user experience of AI

Writing AI in Academic Practice (LinkedIn) · 10 November 2024 · Blog post

AI Trend 1: Multimodal Revolution: Generative AI at the birth of a new paradigm of computer interaction

Writing AI in Academic Practice (LinkedIn) · 3 November 2024 · Blog post

State of AI: 5 trends in generative AI that will shape academic practice in 2025

Writing AI in Academic Practice (LinkedIn) · 1 November 2024 · Blog post

Semantic Machines

Talks Digital Humanities Summer School · 14 August 2024 · Talk

The major breakthrough of generative AI and specifically Large Language Models is the ability to work with unstructured documents with no semantic mark up. Combined with broad cross linguistic capabilities, this has great potential for…

Talk and panel on AI at Tsinghua University

Talks Tsinghua University · 20 March 2024 · Talk

The future of AI in Academic Practice is one million tokens and why it matters

Writing AI in Academic Practice (LinkedIn) · 11 March 2024 · Blog post

2023 8

One year anniversary of ChatGPT this week - where are we, what have we learned?

Writing AI in Academic Practice (LinkedIn) · 27 November 2023 · Blog post

How (not) to learn about AI with metaphors: And how to use ChatGPT as a metaphor generation assistant

Writing AI in Academic Practice (LinkedIn) · 17 August 2023 · Blog post

Have you asked your bot to make you a table? ChatGPT as a cognitive scaffolding engine

Writing AI in Academic Practice (LinkedIn) · 8 August 2023 · Blog post

The State of AI for August 2023: 6 trends from the last 3 months and 6 actions you can take today

Writing AI in Academic Practice (LinkedIn) · 1 August 2023 · Blog post

Invited talk at the European Educational Publishers Group conference

Talks European Educational Publishers Group conference · June 2023 · Talk

Learn to Be a Prompt Engineer: PREP and other guides to uncovering the secrets to ChatGPT mastery

Writing AI in Academic Practice (LinkedIn) · 2 April 2023 · Blog post

More AI news: Incremental but meaningful progress - From models to products

Writing AI in Academic Practice (LinkedIn) · 27 March 2023 · Blog post

ChatGPT is a Weird Linguist: What Should Language Teachers Know?

Talks TILT Webinar (Association for Language Learning) · 2023 · Webinar

In this specialized talk for language professionals, LLMs are described as 'superpolyglots' but 'weird linguists.' While models like GPT-4 can translate between languages with high fluency, their acquisition method (reading the entire…

2022 1

Unintentional Pygmalions: 4 questions to ask when checking an artificial entity for sentience and how to think about the answers

Writing Metaphor Hacker · 9 August 2022 · Blog post

The piece proposes tests for persistent identity, autonomous intentionality, embodiment, theory of mind, and conversational implicature when assessing claims of machine sentience.

2019 1

Turing tests in Chinese rooms: What does it mean for AI to outperform humans

Writing Metaphor Hacker · 23 July 2019 · Blog post

The piece argues that benchmark victories do not establish human-like understanding and proposes prolonged supervised performance as a stronger test of artificial intelligence.

2018 3

3 “easy” things that are hard for both humans and AI

Writing Metaphor Hacker · 19 October 2018 · Blog post

The piece uses time zones, scheduling, and data transfer to show why seemingly simple tasks remain difficult for both people and artificial intelligence.

Does machine learning produce mental representations?

Writing Metaphor Hacker · 28 May 2018 · Blog post

The piece argues that machine-learning systems represent problem spaces without possessing the flexible, embodied, and metacognitive mental representations used by humans.

Learning vs training in machines and organizations: Production of knowledge vs production of capability

Writing Medium · 6 February 2018 · Essay

The essay distinguishes learning as the production of knowledge from training as the production of capability in machines and organisations.

2007 2

Cognition, information, knowledge and the limits of serial computing

Writing Hermeneutic Heretic · 11 September 2007 · Blog post

The piece argues that growing computing power and information volume will not yield human cognition or knowledge without models of embodied and social processes.

Algorithms and everything

Writing Hermeneutic Heretic · 9 January 2007 · Blog post

The piece argues that algorithmic descriptions are useful only at particular scales and must coexist with incompatible perspectives.