“Policy must be based on science, not science fiction.”
— Fei-Fei Li

A personal guide to understanding, using, and questioning artificial intelligence—from computer vision and spatial computing to art, culture, and public policy.
Last updated: October 2026
Artificial intelligence is changing how we make images, write software, conduct research, and organise work. It is also producing an extraordinary amount of noise. Research findings, product announcements, speculation, and advertising often circulate together, with little to distinguish them.
As a visual designer, I want to understand the technologies entering my practice. How does a model interpret an image? What does it learn from its training data? How much control does an artist have over the result? What are the cultural, environmental, and political consequences of the systems we use?
This sourcebook is an attempt to build a useful foundation for those questions. It is also the reading behind my own work: the algorithmic mosaics I have made for more than twenty years depend on the same questions of perception, geometry, and machine vision.
My main interests are computer vision, vision-language models, graphics, 3D reconstruction, spatial AI, and creative computing. But technical understanding needs a wider context. The directory also includes journalism, social research, philosophy, policy, and the history of art and technology.
I have tried to look beyond the institutions and publications that dominate English-language discussion. Research communities in China, India, Africa, Latin America, the Middle East, and elsewhere bring different languages, priorities, and experiences to the field. Geography matters because it shapes which problems receive attention, whose data becomes available, and who benefits from the results.
This is a working directory, shaped by those interests. Inclusion means that a source offers something useful to read, examine, or test. It does not mean agreement with everything it publishes.
Where to begin
The full directory is extensive. These ten resources offer different ways into it:
- Elements of AI — An accessible introduction for readers without a technical background.
- 3Blue1Brown — Visual explanations of the mathematics behind neural networks.
- Foundations of Computer Vision — A substantial introduction to how computers interpret images.
- Google PAIR: AI Explorables — Interactive essays on model behaviour, privacy, and fairness.
- Hugging Face Learn — Practical courses connected to models, code, and demonstrations.
- Simon Willison’s Weblog — Hands-on investigations of what language-model tools can and cannot do.
- AI as Normal Technology — Analysis of capabilities, adoption, and exaggerated claims.
- Stanford AI Index — A broad reference for data about AI research, investment, and development.
- Rest of World — Reporting that brings experiences outside the usual Western technology centres into view.
- Ars Electronica — A place to explore how artists engage with technology and society.
For a practical route, move from an introductory course to a small project and its documentation. For a research route, begin with a textbook before following recent papers. For a critical route, read a technical claim alongside independent reporting and research on its social consequences.
How I read these sources
Different sources answer different questions.
A research paper can explain a method and its experimental results. Documentation can show how to use a tool. Journalism can investigate what happens when a system enters the world. Critical scholarship can examine the assumptions that shaped it.
Before accepting a claim, I find these questions useful:
- What kind of source is this? A reviewed paper, preprint, company announcement, investigation, or opinion essay?
- Who produced it, and who supports the work? Institutional interests do not settle whether a claim is true, but they help explain its framing.
- What was actually tested? A benchmark result, a selected demonstration, and reliable performance in everyday use are different kinds of evidence.
- Can others examine the result? Look for methods, code, data documentation, limitations, and independent evaluation.
- What does “open” mean here? Available code, downloadable model weights, disclosed training data, and permission for commercial use are separate questions.
- What is missing? Languages, communities, failure cases, costs, and working conditions can disappear from a technically impressive account.
Research software may require considerable adaptation. Publicly downloadable datasets may carry restrictions. An archive can remain valuable long after its tools stop receiving updates.
The sections below group sources by their main purpose. Institutions often work across several of them.
1. Scientific and Academic Research
Computer vision, graphics, and spatial understanding
CVPR, ICCV, and ECCV proceedings
Research on image recognition, segmentation, video, reconstruction, and related fields. The Computer Vision Foundation’s open-access archive is a direct route to papers and supplementary material, including the workshops attached to each conference. Workshops often examine narrower or emerging topics; their scope and review arrangements vary by event.
IEEE Transactions on Pattern Analysis and Machine Intelligence
A research journal covering computer vision, pattern recognition, and machine learning. Useful for detailed studies that develop beyond the space available in conference papers.
Stanford Vision and Learning Lab — United States
Research connecting visual perception, learning, and interaction with the physical world. The associated CS231n course provides a structured technical introduction.
Berkeley Artificial Intelligence Research — United States
Research across vision, robotics, learning, and decision-making. The BAIR Blog offers explanations of selected projects alongside their papers.
MIT Computer Science and Artificial Intelligence Laboratory — United States
A broad research institution with work in vision, graphics, robotics, and computational imaging. Follow individual groups and publications for a more focused view.
Tsinghua University Department of Computer Science and Technology — China
An entry point to research groups working on vision, multimedia, machine learning, and related systems. Consult individual laboratories for papers and project code.
Chinese Journal of Computers — China
A journal covering computer science, including AI and visual computing. Primarily a Chinese-language resource; useful for widening the literature beyond English-language venues.
Chinese Conference on Pattern Recognition and Computer Vision: PRCV — China
Proceedings covering pattern recognition, image analysis, and computer vision. A further route into research from Chinese institutions and their collaborators.
Oxford Visual Geometry Group — United Kingdom
Research in visual recognition, geometry, video, and multimodal learning, with a substantial collection of datasets, software, and earlier work.
Inria WILLOW — France
Research on visual learning and the interpretation of images and video, including recognition, human actions, and 3D understanding.
ETH Zürich Computer Vision Lab and Computer Vision and Geometry Group — Switzerland
Distinct groups working across visual learning, reconstruction, localisation, and geometry. Their publications offer complementary approaches to understanding scenes and recovering spatial structure.
Max Planck Institute for Intelligent Systems: Perceiving Systems — Germany
Research on humans in images and 3D environments, including body shape, pose, motion, and interaction. The SMPL model is an important related resource.
Centre for Visual Information Technology, IIIT Hyderabad — India
Research in vision, document understanding, recognition, and related areas, including work relevant to Indian scripts and visual material.
Video Analytics Lab, Indian Institute of Science — India
Research on computer vision, video analysis, and machine learning. Useful for following methods alongside their mathematical foundations.
Indian Conference on Computer Vision, Graphics and Image Processing: ICVGIP — India
A conference connecting vision, graphics, and image processing. Its proceedings provide another view of the field’s research communities and applications.
INSAIT — Bulgaria
A research institute working in computer science and AI, including vision, robotics, language models, and trustworthy systems. Its publication lists provide the most direct view of its work.
KAUST Visual Computing Center — Saudi Arabia
Research across graphics, imaging, computer vision, and visualisation. Relevant to readers interested in the connection between visual computing and scientific applications.
University of Tokyo: Next Generation Artificial Intelligence Research Center — Japan
An interdisciplinary centre studying machine intelligence, embodied interaction, and the relationship between human and artificial intelligence.
Specialised research venues
SIGGRAPH and SIGGRAPH Asia
Conferences connecting graphics research, animation, simulation, imaging, and interactive techniques. Ke-Sen Huang’s paper lists are a useful companion for locating graphics papers.
International Conference on 3D Vision: 3DV
Research focused on recovering and understanding three-dimensional structure from visual information, including depth, reconstruction, and scene analysis.
ACM Multimedia
Research spanning images, video, audio, language, and their combinations. Relevant to understanding systems that work across several media.
ICDAR and IAPR Technical Committee 11
Research on document analysis and recognition: reading text in images, interpreting page layouts, understanding handwriting, and working with historical documents.
IEEE Intelligent Vehicles Symposium
Research on vehicle perception, planning, and automated driving. Useful for seeing how visual methods are evaluated in systems with demanding physical constraints.
IEEE Geoscience and Remote Sensing Society conferences
Including IGARSS, these venues cover satellite imagery, Earth observation, and remote sensing. Relevant to environmental monitoring and the interpretation of large-scale spatial data.
Medical Image Computing and Computer Assisted Intervention: MICCAI
Research on medical imaging and computational methods for healthcare. Publication at a research venue should be distinguished from evidence of clinical readiness.
Broader AI and machine learning
arXiv
A repository of research manuscripts. Useful categories include computer vision, machine learning, AI, and graphics. Posting on arXiv does not itself constitute peer review.
NeurIPS, ICML, and ICLR
Major venues for machine learning research. OpenReview also provides access to submissions, reviews, and discussions for participating venues.
Journal of Machine Learning Research
An open-access journal covering machine learning methods, theory, and applications. A useful source for substantial treatments of a topic.
Science China Information Sciences — China
An English-language journal covering information science, including AI, computing, and related engineering research.
Chinese Academy of Sciences — China
An institutional entry point to research institutes, publications, and scientific developments. Follow the relevant institute or original paper beyond the news summary.
Tsinghua Institute for AI Industry Research — China
Research connecting AI methods with industrial and scientific applications. Useful for examining how academic research moves toward deployment.
ELLIS — Europe
A network of European research units working on machine learning and related disciplines. Its distributed structure makes it useful for discovering groups beyond a few familiar universities.
The Alan Turing Institute — United Kingdom
Research and public material on data science and AI, including methodological, applied, and societal questions.
IndiaAI — India
A government-backed portal covering India’s AI initiatives, research, skills, and infrastructure, including national programmes for access to computing. Read it as an institutional source on national priorities, and check current programme documentation for eligibility.
IIIT-Delhi — India
An entry point to research groups working in AI, language, vision, and related areas. Individual faculty and laboratory pages provide more precise reading lists.
Tata Institute of Fundamental Research — India
A broader source for foundational work in mathematics and computer science, including research relevant to learning theory and computation.
Mila — Canada
A research institute working across machine learning, applications, and responsible AI. Its publications connect foundational methods with interdisciplinary questions.
CSIRO Data61 — Australia
Research in data science, AI, robotics, and digital systems, often connected to scientific and industrial applications.
Masakhane and Deep Learning Indaba — Africa
Distinct initiatives supporting African AI research. Masakhane focuses on African-language NLP; Indaba supports research capacity and community across the continent.

2. Tools, Models, Code, and Datasets
Tools make research tangible. Their documentation, licences, maintenance history, and examples often matter as much as their headline results.
Finding papers and models
Hugging Face Papers
A community-oriented way to discover research and associated projects. Popularity helps identify discussion; it does not establish scientific quality. The former Papers with Code entries are consolidated here.
Hugging Face Model Hub
A repository of models, model cards, and associated files. Inspect each model’s documentation and licence: availability on the platform does not imply uniform openness or reliability.
Hugging Face Tasks
Introductions to tasks such as segmentation, detection, depth estimation, and image-text processing, with links to models and examples.
Open VLM Leaderboard
A resource for comparing vision-language models on selected evaluations. Read the benchmark definitions and testing conditions before interpreting a ranking.
GitHub Trending
A discovery tool for software attracting attention. Stars and rapid growth are signals of interest, rather than substitutes for checking code quality, maintenance, and usefulness.
Computer vision libraries and repositories
PyTorch Hub and torchvision models
Model implementations and pretrained weights for common vision tasks. Useful starting points for experiments, with requirements and limitations that vary by model.
NVIDIA NGC
A catalogue of containers, models, and software aimed at NVIDIA computing environments. Relevant when hardware compatibility and deployment are part of the problem.
NVIDIA 3D Vision research
Research projects on visual understanding, reconstruction, and related 3D methods. Check each project for available code, weights, and usage terms.
ModelScope
A model and development platform with substantial participation from Chinese research and industry. Useful for discovering models and tools beyond the most familiar English-language hubs. Its guides and examples vary in language and completeness between projects.
OpenGVLab
Research code and models in computer vision and multimodal learning, including the InternVL family.
PaddleDetection
A PaddlePaddle-based toolkit for detection, segmentation, tracking, and pose estimation, with models and deployment examples.
OpenVINO Open Model Zoo
Models and demonstrations associated with the OpenVINO ecosystem. Check current compatibility and release guidance before adopting an example.
Roboflow Universe and Supervision
Universe provides community datasets and model projects; Supervision supplies reusable vision utilities. Dataset permissions and annotation quality require individual inspection.
CVIT repositories
Code associated with research at IIIT Hyderabad’s Centre for Visual Information Technology. Useful for connecting a paper’s claims to its implementation.
NAVER Clova AI repositories
Research implementations spanning OCR, document understanding, and other AI topics. Consult individual repositories for their scope and maintenance status.
Kakao Brain repositories
A collection of released research projects, including work on image generation and multimodal learning. Treat each repository according to its own release history.
MBZUAI ORYX repositories
Research code covering visual learning and related methods. Use project documentation to identify datasets, training requirements, and suitable applications.
Vision foundation models
These are the general-purpose models I return to most often in my own image work: they describe, separate, and measure pictures rather than generate them.
OpenCLIP
An open implementation of CLIP-style models that place images and text in a shared space. Useful for searching, sorting, and comparing large image collections by meaning rather than filename.
DINOv2
Self-supervised visual features learned without captions. Useful for measuring similarity between images and for tasks where text labels are a poor description of what matters visually.
SAM 2: Segment Anything
Models for separating objects and regions in images and video from a click, box, or mask. A practical way to see how far automatic segmentation has come, and where it still fails.
Depth Anything
Monocular depth estimation from a single photograph. Relevant to anyone turning flat images into layered or spatial compositions.
Image generation and multimodal systems
arXiv: Multimedia Computing
Preprints involving several media, including video, audio, images, and language. Useful alongside the vision and machine learning categories.
Hugging Face Diffusers
A library for working with diffusion models, including generation, editing, and conditioning workflows. Its documentation connects individual techniques to runnable examples.
ComfyUI
A node-based interface for building image- and video-generation workflows. Its graphs make each step of a generative process visible and repeatable, which matters when an artist wants control rather than a lucky result.
ControlNet and T2I-Adapter
Research introducing additional visual controls into image-generation systems. Useful for understanding how edges, poses, depth, and other conditions can guide an output.
Stability AI research
Research and releases associated with generative models. Read the terms attached to each release rather than assuming a shared licence across the catalogue.
LAION
A research organisation associated with large-scale image-text datasets and open machine learning. Relevant to studying how training collections are assembled and documented.
3D reconstruction, spatial AI, and video
Nerfstudio
A development framework for neural rendering and related reconstruction methods. A practical entry point to experimenting with radiance-field representations.
OpenNeRF
A specific research project on open-vocabulary 3D scene segmentation. Useful for exploring how visual-language features can support searching and interpreting reconstructed scenes.
COLMAP
Software for recovering camera positions and 3D structure from photographs. A useful foundation for understanding the geometry behind many reconstruction workflows.
gsplat
An open-source library for differentiable rendering of 3D Gaussians, with examples and evaluation material. Relevant to practical work with Gaussian splatting.
Something-Something dataset
Video data focused on actions and object interactions. Useful for examining how models distinguish movement and relationships over time.
VideoMAE and TimeSformer
Two research approaches to video representation: learning through masked reconstruction and modelling space-time relationships with attention.
Ego4D
A research dataset and benchmark collection built around first-person video. Relevant to activity understanding, memory, interaction, and embodied perception.
Documents, remote sensing, and medical imaging
PaddleOCR and EasyOCR
Tools for recognising text in images. Evaluate them on the scripts, layouts, image quality, and working conditions relevant to your project.
DocVQA and PubLayNet
Resources for document question answering and page-layout analysis. They address different tasks and should be chosen accordingly.
LayoutLM research and code
Models that combine textual content with document layout and, in later variants, visual information. Relevant to forms, scanned pages, and visually structured documents.
Microsoft Planetary Computer
Access to environmental and geospatial datasets with tools and examples for analysis. Useful for exploring satellite imagery and changes in the physical world.
EuroSAT and BigEarthNet
Remote-sensing datasets used in land-cover research. Check their versions, labels, and evaluation conventions when comparing results.
Radiant Earth
An organisation working on Earth observation, open geospatial data, and machine learning. Relevant to questions of access and public benefit in geospatial technology.
MONAI
An open-source framework for medical imaging AI, including data processing, training, and evaluation components.
Medical Segmentation Decathlon
A collection of medical-image segmentation tasks for evaluating methods across different datasets. Benchmark performance alone does not establish suitability for patient care.
NIH Clinical Center
An institutional starting point for clinical research and related data resources. Follow the specific study or dataset for access conditions and documentation.
Robotics and mobility
AI Habitat
Simulation tools and research resources for embodied AI, including navigation and interaction in virtual environments.
Open X-Embodiment
A collaborative effort bringing together robot-learning datasets across different platforms. Useful for studying what can transfer between robots and tasks.
nuScenes and Waymo Open Dataset
Datasets supporting autonomous-driving research, including perception and tracking. Sensor configurations, geography, and recording conditions shape what their results can tell us.
Baidu Apollo
An autonomous-driving development platform. Relevant to understanding how perception, planning, and control fit together within a larger system.
SAE J3016
A reference taxonomy for levels of driving automation. Relevant when interpreting claims about autonomous vehicles; the level labels are not a general measure of system safety.
Infrastructure and model ecosystems
MLflow
Tools for tracking experiments and managing models and related development workflows. Useful when reproducibility and project organisation become difficult to manage manually.
Weights & Biases
Tools and community material for recording experiments, comparing results, and examining training behaviour, now part of CoreWeave. Public reports range from tutorials to project-specific analyses.
LiteRT and ExecuTorch
Deployment frameworks for running models on constrained devices. These are the current directions to consult when encountering older TensorFlow Lite or PyTorch Mobile material.
tinyML
A community focused on machine learning in devices with limited computing power, memory, and energy. Useful for understanding the constraints hidden by large-model demonstrations.
Qwen, DeepSeek, Z.ai, and Moonshot AI
Official model repositories from Chinese AI developers. Compare documentation, licences, available files, and independent evaluations for each release.
OpenI
A platform for AI development and collaboration, including code, models, and computing resources. Useful for discovering another part of the Chinese development ecosystem.
Mistral AI, EuroLLM, and OpenEuroLLM
Distinct European model efforts with different organisations, goals, and release arrangements. Relevant to multilingual access and the development of European AI capabilities.
EuroHPC Joint Undertaking
Institutional information on European supercomputing and AI infrastructure, including AI Factories. Useful for understanding access programmes and public investment.
ONNX Model Zoo — Historical resource
The repository is preserved for historical purposes and points readers to Hugging Face for its models. Distinguish this collection from the ONNX model format itself.
iHub-Data, IIIT Hyderabad
An initiative connecting data-driven research, training, and applications. Relevant to the relationship between academic work and practical development.
MBZUAI Institute of Foundation Models and LLM360
Sources for foundation-model research and releases with an emphasis on sharing development artefacts. Examine each release for the materials actually provided.

3. Critical, Philosophical, and Social Perspectives
Understanding a system also means examining the institutions, assumptions, and labour behind it.
Critical AI
An interdisciplinary research initiative associated with the journal Critical AI. Useful for cultural, historical, and political approaches to AI.
AI Now Institute
Research and policy analysis on corporate power, accountability, and the institutions shaping AI development.
Data & Society
Research on the social consequences of data and automated systems, including labour, inequality, governance, and public life.
AlgorithmWatch
Research, investigations, and advocacy concerning automated decision-making and its effects on rights and democratic institutions.
Ada Lovelace Institute
Research on the social implications and governance of data and AI, with attention to public participation and the distribution of benefits and harms.
Distributed AI Research Institute: DAIR
Research that places communities and the conditions of AI development at the centre of the inquiry. Useful for questioning assumptions embedded in mainstream research agendas.
UNESCO: Ethics of AI
International principles and implementation resources concerning AI ethics. Read alongside evidence of how those principles are applied in particular settings.
Centre for Internet and Society — India
Research on digital rights, governance, access, and technology, bringing Indian and wider Global South perspectives to the discussion.
ACM FAccT
An interdisciplinary conference on fairness, accountability, and transparency. Its research examines both technical methods and the settings in which systems operate.
Stanford Human-Centered AI
Research and public material on AI’s development and consequences, from technical work to policy and education.
Big Data & Society
An open-access journal examining the social, cultural, and political dimensions of data-intensive systems.
Excavating AI — Kate Crawford and Trevor Paglen
An essay on the images and labels inside training sets, and the politics of classifying people and pictures. Essential reading for anyone who works with images.
Knowing Machines
A research project tracing the histories, practices, and politics of how machine learning systems are trained, with close studies of large image datasets.
Algorithmic Justice League
An organisation combining research, art, education, and advocacy to examine AI harms. Particularly relevant to the relationship between visual systems and representation.
Derechos Digitales — Latin America
Research and advocacy on digital rights, surveillance, and technological power. An important perspective for understanding how AI intersects with regional political and social conditions.
4. Geometry, Graphics Standards, and Computing Foundations
Many tools used in visual practice depend on ideas that predate the current AI boom: geometry, graphics standards, numerical methods, and efficient computation.
W3C SVG Working Group
Standards and technical material for Scalable Vector Graphics. Relevant to anyone working with paths, coordinates, typography, and graphics on the web.
The Khronos Group
A consortium developing standards for graphics and computing, including Vulkan, glTF, and OpenVX. Useful for understanding how visual systems exchange data and use hardware.
Eurographics
A European research community covering modelling, rendering, animation, and geometry processing. With the SIGGRAPH archive, it places recent neural methods within a longer graphics tradition.
CGAL
A computational geometry library with implementations of triangulations, meshes, spatial structures, and other geometric operations. Relevant when a project requires dependable geometry.
Open CASCADE Technology
An open-source platform for 3D modelling and CAD development, including boundary representations and geometric operations.

5. Related Fields: Language, Audio, Causality, and Safety
Vision increasingly intersects with other forms of information and action. These resources help make those connections clearer.
ACL Anthology
An open archive of computational linguistics and natural language processing research, including ACL and EMNLP proceedings. Useful for multilingual research and the language components of multimodal systems.
ISCA and Interspeech
Research communities and publications covering speech processing, recognition, synthesis, and spoken interaction.
LibriSpeech
A widely used speech-recognition dataset based on read audiobooks. Its recording conditions help define what evaluations using it can—and cannot—demonstrate.
Whisper
An open release for speech recognition and translation, with code and model information. Test performance on the languages, accents, and recording conditions that matter to your work.
Conference on Robot Learning: CoRL
Research connecting machine learning and robotics, including perception, control, manipulation, and learning from demonstrations.
Ray RLlib
Software for reinforcement learning experiments and applications. Relevant to the practical study of systems that learn through interaction.
Google DeepMind research
Research across learning, reasoning, scientific applications, and robotics, including systems that connect perception, language, and action. Read project announcements alongside the underlying methods and evaluations.
Conference on Causal Learning and Reasoning: CLeaR
Research on causal inference and its relationship to machine learning. Useful for distinguishing prediction from claims about cause and effect.
DoWhy
A library for causal modelling and analysis, including ways to examine assumptions and test the robustness of an inference.
UCLA Cognitive Systems Laboratory
Research and educational material associated with Judea Pearl and causal reasoning.
Center for Human-Compatible AI
Research on AI systems that account for human preferences and uncertainty about those preferences.
AI Alignment Forum
A technical discussion forum on alignment questions. Read it as a research community with particular assumptions and priorities, rather than a survey of the entire field.
Anthropic research
Company research on model behaviour, interpretability, evaluation, and safety. Relevant technical work that should also be read in its commercial context.
UK AI Security Institute
Government research and evaluation concerning advanced AI capabilities and risks. Useful for following institutional approaches to testing systems.

Faces of Innovation in Academic Program Development, for Communications of the ACM.
6. Learning: Books, Courses, and Visual Explanations
A durable understanding comes from returning to foundations. These resources offer different balances of mathematics, explanation, and practical work.
Foundations of Computer Vision — Antonio Torralba, Phillip Isola, and William Freeman
An online textbook covering image formation, geometry, visual processing, and learning. Particularly relevant to readers whose interest begins with images.
Computer Vision: Algorithms and Applications — Richard Szeliski
A broad reference covering the methods and applications of computer vision, with material available through the author’s website.
Multiple View Geometry in Computer Vision — Richard Hartley and Andrew Zisserman
A mathematical foundation for understanding cameras, perspective, stereo, and reconstruction. Best approached with some preparation in linear algebra.
Understanding Deep Learning — Simon J. D. Prince
A textbook combining explanations, diagrams, and accompanying computational material. Useful for developing a coherent view of modern neural networks.
Deep Learning — Ian Goodfellow, Yoshua Bengio, and Aaron Courville
A foundational reference on neural networks and their mathematical background. Its older publication date makes it a base for later reading rather than a guide to every current architecture.
Dive into Deep Learning
A textbook that connects concepts to executable code. A Chinese edition offers another route through the material.
MIT 6.S191: Introduction to Deep Learning
Lectures and practical material introducing deep learning methods and applications. Consult the edition relevant to the material you are studying.
Practical Deep Learning for Coders — fast.ai
A practical course that begins with working applications and develops the underlying ideas through implementation.
Elements of AI
An introduction for readers without a programming background, covering basic ideas and questions about AI’s use.
Hugging Face Learn
Courses connected to current open development tools, including language models, vision, diffusion, and agents.
3Blue1Brown
Animated explanations of mathematics, including neural networks and transformers. Useful for building visual intuition before or alongside formal study.
Neural Networks: Zero to Hero — Andrej Karpathy
A programming-based course that constructs neural networks step by step. Requires a willingness to work directly with code.
The Illustrated Transformer — Jay Alammar
A visual introduction to the transformer architecture. A helpful starting point before reading more recent variations.
Distill — Archive
Interactive research articles exploring neural networks and representation. The archive remains valuable for both its explanations and its approach to scientific communication.
Google PAIR: AI Explorables
Interactive essays that make questions about fairness, privacy, representation, and model behaviour easier to investigate. Produced within Google’s research ecosystem.

Does the internet make you smarter? For The Wall Street Journal, with an essay by Clay Shirky.
7. Newsletters, Blogs, and Journalism
These sources help connect individual papers and announcements to a wider conversation. Their strengths differ: explanation, experimentation, reporting, criticism, or informed commentary.
Import AI — Jack Clark
Research summaries and commentary on AI’s technical and political development. Read with awareness of the author’s industry involvement.
Ahead of AI — Sebastian Raschka
Technical explanations of machine learning research, often supported by diagrams, implementation details, and comparisons.
Simon Willison’s Weblog
Practical investigations of language models and software tools, including experiments, unexpected behaviour, and security concerns.
Lil’Log — Lilian Weng
Long-form technical surveys of topics such as generative models, agents, and model behaviour. Particularly useful as maps to further reading.
The Gradient
Essays and interviews connecting AI research with its history, practice, and social implications.
Interconnects — Nathan Lambert
Analysis of model development, post-training, reinforcement learning, and the open-model ecosystem, written from a research practitioner’s perspective.
ChinAI — Jeffrey Ding
Translations and analysis that help English-language readers follow Chinese discussions of AI.
Synced
AI research and industry coverage, including work from China. Follow links to original papers when evaluating technical claims.
AI as Normal Technology — Arvind Narayanan and Sayash Kapoor
Analysis of AI capabilities, adoption, and institutions, continuing the authors’ work on exaggerated or misleading claims.
Marcus on AI — Gary Marcus
Critical commentary on current systems and claims about their future. Useful as an explicitly argued position to compare with competing accounts.
MIT Technology Review: AI
Journalism examining research, companies, policy, and the people affected by technology. Some material requires a subscription.
Rest of World
Reporting on technology beyond the most familiar Western centres, including its effects on workers, businesses, and communities.
Two Minute Papers
Accessible video introductions to graphics, simulation, and AI papers. Treat the videos as invitations to inspect the research, especially when demonstrations are striking.
AI Weirdness — Janelle Shane
Playful experiments that make model behaviour and failure memorable. The AI archive is particularly useful for resisting overly human interpretations of generated output.
404 Media
Independent investigative technology journalism, including reporting on AI products, data practices, surveillance, and their consequences.
Understanding AI — Timothy B. Lee
Reporting and explanation connecting AI’s technical workings to economic and social questions. A useful bridge between research and public discussion.

8. Art, Design, and Creative Computing
Artists have worked with computation for decades. That history helps put generative AI into perspective and opens questions that product demonstrations rarely address.
Ars Electronica — Austria
Exhibitions, a festival, a prize, and research connecting art, technology, and society. Useful for exploring both artistic possibilities and critical responses.
ZKM | Center for Art and Media — Germany
Exhibitions, collections, and research on media art and technological culture, including the histories surrounding contemporary digital practice.
Victoria and Albert Museum — United Kingdom
Collections and research documenting computer art and digital objects. Useful for seeing present tools within a longer history of artistic experimentation.
Computer Arts Society
A community and archive supporting the study and practice of computer art. A valuable companion to museum collections and individual artists’ archives.
Leonardo
A journal exploring art, science, and technology through research and artists’ accounts of their practice.
ISEA International
An international organisation associated with the Symposium on Electronic Art. Its events and archives reveal a wide range of approaches to technologically mediated art.
Processing Foundation and p5.js
Tools and educational communities supporting creative coding. Relevant to artists who want to construct their own systems and understand how an image is produced.
Serpentine Arts Technologies
Artistic commissions and research exploring emerging technologies and their cultural implications.
Rhizome
Art, writing, and preservation work concerned with digital culture and born-digital practice.
CreativeApplications.Net
Projects and commentary connecting art, design, computation, and interaction. Useful for discovering practices beyond the familiar categories of commercial software.
Gray Area
An arts and technology organisation with exhibitions, education, and community programmes.
Lumen Prize
A prize and platform for art made with technology. Useful for discovering artists and comparing how different practices use computational tools.
AIxDESIGN
A community bringing design, art, and critical inquiry to AI. Relevant to questions of participation, representation, and alternative approaches to making systems.
Rights, consent, and the artist’s position
Glaze and Nightshade
Tools from the University of Chicago that alter images to resist style imitation and unauthorised training. Useful both as protection and as a lesson in how fragile model perception can be.
Creative Commons: AI and the Commons
Work on how openly licensed culture, artists’ preferences, and AI training can coexist. A reference point for the question every working artist now faces: what happens to my images once they are online.

I am your watchdog! For a Wall Street Journal article on information security.
9. Policy, Public Data, and Evidence
Policy sources explain how institutions respond to AI. Data sources help test claims about its growth and impact. Both need attention to definitions, methods, and dates.
Governance and public policy
EU AI Act: official text
The primary legal text of Regulation (EU) 2024/1689. The AI Act Explorer provides a separate, accessible way to navigate its provisions.
Council of Europe: AI Framework Convention
Official information on the convention concerning AI, human rights, democracy, and the rule of law. Consult current materials for participation and implementation developments.
OECD.AI
Policy resources, indicators, and comparative information about national AI developments. Useful when examining how countries define and pursue their priorities.
NIST AI Risk Management Framework
A voluntary framework and associated resources for thinking systematically about AI risks and their management.
DigiChina
Translations and analysis of Chinese technology policy. Useful for examining source documents that are otherwise difficult to access in English.
Center for Security and Emerging Technology: CSET
Research and data on AI, security, talent, and technological development, including translations of Chinese documents. Its security-policy focus informs its selection of questions.
African Union: Continental AI Strategy
Official material on the African Union’s shared approach to AI development and governance. Read alongside national policies and research from African communities.
Centre for the Governance of AI
Research on institutions, policy choices, and the governance of advanced AI systems.
Data and evidence
Stanford AI Index
An annual reference compiling evidence on AI research, investment, capabilities, and other developments. Consult its underlying sources and definitions when using a statistic.
Epoch AI
Research and datasets on model development, computing resources, and related trends. Useful for examining the evidence behind claims about scale.
Our World in Data: Artificial Intelligence
Charts and explanations drawing together data on AI development. Follow the source notes when comparing periods or interpreting a trend.
State of AI Report
A periodic synthesis of research, industry, policy, and safety developments. Useful as an overview, with editorial judgments that should be read as such.
World Bank Open Data and India Data Portal
Public development, social, and economic data. General resources rather than AI sources, useful for testing claims about who benefits from AI and where.
NASSCOM
Industry perspectives and reports on technology adoption in India. Useful alongside academic and public-interest accounts of the same developments.
UNDP Digital and Digital Public Goods Alliance
Distinct initiatives relevant to public-interest digital development. Useful for examining how software, data, and institutional capacity connect to development goals.
AI Incident Database
A public record of harms and near-harms involving deployed AI systems. A useful counterweight to demonstrations: evidence of what happens after release.
Documentation, provenance, and verification
Datasheets for Datasets
A proposal for documenting why datasets were created, how they were assembled, and how they should be used. A useful checklist when examining training material.
Model Cards for Model Reporting
A framework for documenting intended uses, evaluation, performance differences, and limitations. Helpful when deciding whether a model is appropriate for a particular task.
C2PA and Content Credentials
A technical standard for recording media provenance and editing history. Relevant to creative attribution and transparency; a provenance record does not by itself establish the truth of a depicted event.
Bellingcat resources
Methods and examples for investigating images, video, locations, and online evidence. Useful for grounding discussions of synthetic media in practical verification work.
10. Greece
Greek research deserves a visible place in this directory, both for its contribution to the field and for work involving the Greek language, local institutions, and cultural material.
Archimedes Research Unit, Athena Research Center
Research in AI, data science, and algorithms, connecting foundational questions with applications.
Institute for Language and Speech Processing
Research on language and speech technologies, including resources for Greek. Its work is particularly relevant to the relationship between language access and AI development.
Meltemi and Krikri
Open Greek-language models released by the Institute for Language and Speech Processing. A concrete example of a smaller language community building its own models instead of waiting for global ones to serve it well.
Computational Vision and Robotics Laboratory, FORTH-ICS
Research in computer vision and robotics at the Foundation for Research and Technology–Hellas, including perception and interaction with the physical world.
Computer Vision, Speech Communication and Signal Processing Group, NTUA
Research connecting visual information, speech, and signal processing at the National Technical University of Athens.
Institute of Informatics and Telecommunications, NCSR Demokritos
Research across AI, language technologies, data, and computational systems.
Hellenic Artificial Intelligence Society
A scientific society connecting the Greek AI community through research activities, events, and education.
GRNET
Greece’s national research and education network and computing infrastructure, behind much of the country’s public research computing.
AI Factory Pharos
Greece’s AI Factory initiative within the EuroHPC ecosystem, with information on infrastructure, services, and areas of application.
11. Italy
Italy combines a strong computer vision tradition with new national investment in AI research and computing, and its own work on Italian-language models.
FAIR: Future Artificial Intelligence Research
A national research partnership connecting universities, research centres, and companies across Italy. A useful map of where Italian AI research is being funded and organised.
Istituto Italiano di Tecnologia
A research institute in Genoa working across robotics, computer vision, and embodied intelligence, known for the iCub humanoid robot.
AImageLab, University of Modena and Reggio Emilia
A computer vision and multimedia research group working on video understanding, vision-language models, and applications in cultural heritage, among other areas.
CINI AIIS National Laboratory
The Italian national laboratory for Artificial Intelligence and Intelligent Systems, bringing together university research groups across the country.
CVPL
The Italian association for computer vision, pattern recognition, and machine learning, which organises the ICIAP conference series.
AIxIA: Italian Association for Artificial Intelligence
A scientific society connecting the Italian AI research community through conferences, working groups, and education.
Minerva
A family of language models from Sapienza University of Rome’s NLP group, trained with Italian as a primary language. A parallel to the Greek models above.
IT4LIA AI Factory
Italy’s AI Factory within the EuroHPC network, hosted by CINECA in Bologna and built around the Leonardo supercomputer.
12. Cyprus
Cyprus is a small research community with an outsized interest in visual computing, Earth observation, and the eastern Mediterranean. I live and work here, so these are also my neighbours.
CYENS Centre of Excellence
A research centre in Nicosia focused on interactive media, smart systems, and emerging technologies, with work in visual sciences, human factors and design, and artificial intelligence.
KIOS Research and Innovation Center of Excellence, University of Cyprus
Research on intelligent monitoring, control, and security of critical infrastructure, including power and water systems, emergency response, and a digital twin of Cyprus.
ERATOSTHENES Centre of Excellence
An Earth observation and space technology centre hosted by the Cyprus University of Technology in Limassol. Relevant to satellite imagery and remote sensing in the eastern Mediterranean.
The Cyprus Institute: CaSToRC
The Computation-based Science and Technology Research Center, home to the country’s high-performance computing and data science research.
Deputy Ministry of Research, Innovation and Digital Policy
The government body responsible for Cyprus’s research, innovation, and digital policy, including its national approach to AI. Read it as an institutional source on national priorities.
NeMe
A non-profit cultural organisation in Cyprus working at the intersection of contemporary theory and the arts. A local counterpart to the art and technology sources above.
From my own practice
These sources feed directly into my work with algorithmic mosaics, where images are built from thousands of other images, letters, or shapes. If that intersection interests you, these pages show where the reading leads:
- Algorithmic Mosaic — what the practice is and how it differs from traditional photomosaics.
- A History of Mosaic Algorithms — from early tiling research to today’s tools.
- The Mathematics of Mosaic and Mosaic and Perception — the geometry and vision science behind fragmented images.
Keeping the sourcebook useful
The challenge is to maintain a directory that helps people find their bearings as the field changes.
That means revisiting descriptions, replacing broken links, identifying archives, and paying attention to perspectives that remain underrepresented. It also means being willing to remove a resource when its contribution no longer justifies its place.
Suggestions are welcome—especially sources that explain their methods, acknowledge their limits, and help readers ask better questions.