AI ABOVE THE CUT

Resources

AI Learning Resources

A curated path through the tools, guides, studies, and reports worth knowing.

Last reviewed September 7, 2026

The internet does not have an AI-information shortage. It has a sorting problem.

This collection is for people who want to become capable users of AI without following every product launch or mistaking every confident claim for settled evidence. It begins with a short learning path. The larger library then lets you go deeper into practical use, the mechanics of generative AI, research on how it changes work and thinking, workforce implications, and responsible use.

This is a companion to the Get to Know AI field guide. The field guide teaches a way of working. This page points outward to people and institutions that can help you keep learning.

Product features, model names, prices, and usage limits change quickly. The product links below point to official pages rather than freezing those details here.

Brand-new to AI? Start here.

Pick a free chat tool, try your first message, and learn what to do with the answer. No experience needed.

Your first AI chat →

Start here: a 12-link learning path

If you do not want sixty-eight choices, use these twelve in order.

  1. Learn the basic mental model. The University of Helsinki's Elements of AI is a free, approachable introduction for nontechnical learners. Free course
  2. See what a modern assistant can do. OpenAI Academy's Getting Started with ChatGPT covers the interface, first prompts, and common uses. Official guide
  3. Learn to brief the work. Anthropic's prompting best practices emphasize clarity, context, examples, and structure. The principles travel across products. Official guide
  4. Try the same real task in two tools. Use ChatGPT, Claude, or Gemini and compare the returns. Do not stage a trivia contest; use work you understand well enough to judge.
  5. Learn the central limitation. Read the field guide's sections on plausibility and context and verification, then use the NIST Generative AI Profile to see how those risks appear at system level. Government guidance
  6. Study one strong productivity result carefully. Generative AI at Work reports an average productivity gain in a particular customer-support setting and shows that gains varied by worker experience. Working paper
  7. Study the boundary of competence. The Jagged Technological Frontier study shows why performance can improve on tasks inside a model's capabilities and deteriorate outside them. Field experiment
  8. Study the human side of verification. Microsoft's Critical Thinking in AI-Assisted Knowledge Work examines how knowledge workers say AI changes where they apply critical thinking. Treat it as self-reported evidence, not proof that AI makes people less intelligent. Vendor research
  9. See the broader field. The Stanford AI Index 2026 assembles data on capabilities, economics, adoption, policy, education, and public opinion. Independent index
  10. Put workforce claims in perspective. The ILO's Generative AI and Jobs distinguishes occupational exposure from inevitable job loss and emphasizes transformation as well as automation. Intergovernmental
  11. Adopt a risk vocabulary. The NIST AI Risk Management Framework organizes responsible practice around governing, mapping, measuring, and managing risk. Government framework
  12. Keep learning without chasing everything. Subscribe to one thoughtful digest, such as Stanford HAI's research and policy updates, and revisit this library quarterly.

How to read the labels

Every entry below carries a tag. The tag tells you what kind of claim the source can support, and what it cannot.

Official guide
The organization is explaining its own product or teaching material. It is the best source for how the product works today, but not a neutral comparison.
Academic study
A research paper or working paper with a defined method. Read the task, sample, model, and outcome before generalizing the finding.
Government or intergovernmental
Useful for risk, policy, and workforce framing. It may move more slowly than the products.
Vendor research
May contain strong data and careful analysis, but the publisher also has an interest in AI adoption or in its own product.
Consultancy survey
Useful for seeing what leaders report and how organizations are responding. It is not the same as observed performance.
Independent index or analysis
Synthesizes a field. Its value depends on transparent sources and definitions.

Section 1

Choose and try an AI assistant

These are starting points, not endorsements. Begin with a free account and a low-risk task you know well. Compare tools on usefulness, sources, privacy controls, workflow fit, and how easily you can inspect the work.

  1. 01

    ChatGPT Product

    A general-purpose assistant for conversation, writing, analysis, files, images, research, and larger work.

  2. 02

    ChatGPT plans Official product info

    The official comparison of individual and organizational plans. Check this page for current features and limits.

  3. 03

    Claude Product

    Anthropic's general-purpose assistant, often used for writing, analysis, documents, and coding.

  4. 04

    Choosing a Claude plan Official product info

    Anthropic's current plan guide for individual and organizational users.

  5. 05

    Perplexity Product

    A research-oriented answer engine designed to return web-grounded answers with citations. Always open the cited sources; the presence of citations does not guarantee that a claim is supported.

  6. 06

    Choosing a Perplexity plan Official product info

    The official overview of free, individual, education, enterprise, and developer options.

  7. 07

    Gemini Product

    Google's general-purpose AI assistant, with integrations across parts of the Google ecosystem.

  8. 08

    Google AI plans Official product info

    The current comparison of free and paid Google AI access, including Gemini and NotebookLM benefits.

  9. 09

    Microsoft Copilot Product

    Microsoft's consumer AI assistant. Organizations using Microsoft 365 should separately review the product, license, data, and administrative terms that apply to their environment.

  10. 10

    NotebookLM Product

    A source-grounded research and learning tool that works from material you provide. It is especially useful for exploring a bounded document collection, but its synthesis still requires review.

Section 2

Learn practical AI use

Guides and courses that teach the craft: giving a model clear tasks, context, constraints, and examples, then judging what comes back.

  1. 11

    Using ChatGPT Official guide

    A navigable OpenAI Academy hub covering core skills, files, research, personalization, projects, and workflows.

  2. 12

    OpenAI Academy courses Course catalog

    Free self-paced pathways in AI foundations, applied practice, and agents and workflows.

  3. 13

    Prompting Official guide

    A practical OpenAI Academy introduction to giving a model clear tasks, context, constraints, and examples.

  4. 14

    ChatGPT 101 resource guide Official guide

    A beginner-oriented collection of demonstrations and next steps.

  5. 15

    Anthropic Learn Official guide

    Anthropic's collection of guides and courses for using Claude.

  6. 16

    AI Fluency Official course

    Anthropic's learning program for collaborating with AI more deliberately.

  7. 17

    Prompt design strategies Official guide

    Google's concise guide to clear instructions, examples, context, structure, and iteration. It is written for Google's platform, but much of the advice is portable.

  8. 18

    Google AI literacy Learning hub

    Google's learning collection on understanding and using AI.

  9. 19

    Generative AI for Beginners Open course

    A substantial open course from Microsoft with lessons and examples. Some modules are technical; nontechnical readers can use the conceptual sections.

  10. 20

    AI for Beginners Open course

    Microsoft's broader curriculum on artificial intelligence beyond generative chat tools.

  11. 21

    Introduction to generative AI and agents Official course

    A short Microsoft Learn module on language models, generative AI applications, and agents.

  12. 22

    Elements of AI University course

    A free course created by the University of Helsinki and MinnaLearn for people without a technical background.

  13. 23

    Building AI University course

    A follow-on course that moves from AI concepts toward building and evaluating applications.

  14. 24

    Generative AI for Everyone Independent course

    Andrew Ng's nontechnical course on capabilities, use cases, project lifecycles, and social implications.

Section 3

Understand what is under the hood

You do not need to become a machine-learning engineer. You do need enough of a mental model to understand why fluent output can be wrong, why context matters, and why one task can work brilliantly while a similar-looking task fails.

  1. 25

    MIT: Foundation Models and Generative AI University course

    A nontechnical lecture series covering the path to foundation models and their practical implications.

  2. 26

    MIT: Fundamentals of Large Language Models University course

    A deeper, paid self-paced course on LLMs, multimodality, prompting, limitations, and responsible use.

  3. 27

    What are large language models? Technical primer

    Google's compact explanation of tokens, parameters, prompting, and common model behavior.

  4. 28

    Introduction to Large Language Models Official course

    A more structured Google Machine Learning Crash Course module.

  5. 29

    Attention Is All You Need Research paper

    The foundational 2017 paper that introduced the Transformer architecture. This is primary technical literature, not a beginner guide.

  6. 30

    The Illustrated Transformer Independent explainer

    Jay Alammar's visual explanation of the Transformer. It remains one of the clearest bridges between a general mental model and the technical architecture.

  7. 31

    The Illustrated GPT-2 Independent explainer

    A visual explanation of autoregressive language generation: how a model builds a sequence one token at a time.

Section 4

Study AI, productivity, judgment, and creativity

This is where careful reading matters most. A study can be rigorous and still be narrow. Ask four questions every time: Who participated? What task did they perform? Which model and interface did they use? What outcome was actually measured?

  1. 32

    Generative AI at Work Working paper

    A study of 5,179 customer-support agents that reported higher productivity on average, with larger gains for less-experienced workers. Strong evidence for this setting; not a universal productivity number.

  2. 33

    The Jagged Technological Frontier study Field experiment

    A field experiment with BCG consultants showing gains on tasks inside the model's capability boundary and worse performance on a task outside it.

  3. 34

    Experimental Evidence on the Productivity Effects of Generative AI Peer-reviewed

    A controlled study of midlevel professional writing tasks that found faster completion and higher evaluator-rated quality among participants using ChatGPT.

  4. 35

    Critical Thinking in AI-Assisted Knowledge Work Vendor research

    A survey of 319 knowledge workers about when and how they apply critical thinking with generative AI. It measures reported behavior and perceptions, not long-term cognitive change.

  5. 36

    Rethinking AI in Knowledge Work Vendor analysis

    A useful argument for designing AI as a tool for thought, sensemaking, and reflection rather than only an answer machine.

  6. 37

    Tools for Thought Research program

    Microsoft's research program on systems that support human cognition and agency.

  7. 38

    The New Future of Work Research hub

    A long-running Microsoft Research initiative assembling studies on AI, work, organizations, and education.

  8. 39

    AI-Generated “Workslop” Is Destroying Productivity Survey and commentary

    The article that popularized “workslop”: plausible-looking AI output that pushes interpretation and repair onto someone else. The underlying findings are survey-based and self-reported; use the concept, not the headline as a measured universal rate.

  9. 40

    The hidden cost of workslop Vendor research

    The underlying BetterUp Labs and Stanford Social Media Lab report. Useful for method and question wording when evaluating the workslop claims.

  10. 41

    AI assistance and coding skills Vendor research

    A preliminary randomized study of 52 mostly junior software developers that found a learning cost under the tested conditions. Narrow evidence, important question.

  11. 42

    How AI assistance impacts the formation of coding expertise Vendor research

    Analysis of Claude Code sessions exploring planning, execution, verification, and the role of prior expertise. The evidence is product-specific telemetry.

  12. 43

    Economic Index: Learning curves Vendor research

    Anthropic's analysis of how patterns of Claude use change with experience. Useful for studying adoption behavior, with the limits of one product's data.

  13. 44

    Generative AI enhances individual creativity but reduces collective diversity Peer-reviewed

    An experiment using short stories that found higher creativity ratings for individuals receiving AI help alongside lower diversity across outputs. Do not automatically generalize from short fiction to all creative work.

  14. 45

    When and why people think worse of others who use AI Peer-reviewed

    Research on how observers evaluate people who disclose AI use. Helpful for understanding the social layer of adoption, not just task performance.

Section 5

Follow AI at work, in organizations, and across the economy

These reports answer different questions and use different methods. Read them side by side. Workforce exposure is not the same as job loss. Adoption is not the same as value. A survey of executive expectations is not an audited financial result.

  1. 46

    Microsoft Work Trend Index Research hub

    Microsoft's hub for recurring studies on AI and work. It combines surveys, product telemetry, and interpretation; each edition should be read for its specific method.

  2. 47

    2026 Work Trend Index Vendor research

    Microsoft's current report on agents, human agency, and organizational adoption, based on a survey of AI-using workers across ten markets and other Microsoft data.

  3. 48

    The State of AI Consultancy survey

    McKinsey's 2025 global survey on organizational adoption, scaling, and reported business impact. Useful as a management snapshot; responses are self-reported.

  4. 49

    How organizations are rewiring to capture value Consultancy survey

    McKinsey's earlier 2025 report on workflows, governance, talent, and operating changes associated with reported value.

  5. 50

    One year of agentic AI: six lessons Consultancy analysis

    Practitioner lessons on moving agentic systems from demos into operations. Read as field perspective rather than independent proof.

  6. 51

    PwC 2026 Global AI Jobs Barometer Consultancy analysis

    PwC's analysis of job postings, skill change, wages, and productivity across AI-exposed industries. Check definitions before translating exposure into displacement.

  7. 52

    WEF Future of Jobs Report 2025 Employer survey

    A global employer survey on expected job creation, displacement, skill needs, and transformation through 2030. It reports expectations, not guaranteed outcomes.

  8. 53

    BCG AI at Work 2025 Consultancy survey

    A survey of more than 10,000 workers and leaders on use, training, trust, and the gap between access and effective adoption.

  9. 54

    Stanford AI Index 2026 Independent index

    A wide-ranging annual compilation of data on technical performance, investment, business use, policy, education, and public attitudes.

  10. 55

    ILO: Generative AI and Jobs Intergovernmental

    A global occupational-exposure index that stresses the difference between possible automation and likely job transformation.

  11. 56

    IMF: Gen-AI and the Future of Work Intergovernmental

    An analysis of global labor-market exposure, complementarity, inequality, and policy choices.

  12. 57

    Working with AI: measuring occupational implications Vendor research

    A Microsoft Research paper using Copilot conversations to estimate where AI is being used successfully across work activities. It measures observed product interactions, not which occupations will disappear.

Section 6

Use AI responsibly and manage risk

A shared vocabulary for risk, and the practical security guidance that goes with it.

  1. 58

    NIST AI Risk Management Framework Government framework

    A voluntary framework for governing, mapping, measuring, and managing AI risk. A strong common language for organizations of many sizes.

  2. 59

    NIST Generative AI Profile Government guidance

    A companion profile that applies the AI RMF to generative AI risks such as confabulation, privacy, information integrity, bias, security, and human overreliance.

  3. 60

    NIST AI Resource Center Resource hub

    Playbooks, crosswalks, profiles, and implementation resources for the AI RMF.

  4. 61

    OWASP Top 10 for LLM Applications Open security guidance

    A practical map of common application-security risks, including prompt injection, sensitive-information disclosure, excessive agency, and unbounded consumption.

  5. 62

    CISA: Stay Safe Online When Using AI Government guidance

    A short, plain-language guide to protecting private information and recognizing AI-enabled fraud.

  6. 63

    OECD AI Principles Intergovernmental

    International principles covering human rights, fairness, transparency, system reliability, accountability, inclusive growth, and preparation for labor-market change.

  7. 64

    OECD Framework for the Classification of AI Systems Intergovernmental

    A way to characterize an AI system by its context, data and inputs, model, task, output, and effects on people and the planet.

Section 7

Keep up without drowning in AI news

Choose one or two. A healthy information diet combines research, practical interpretation, and reporting. It does not require consuming every launch thread.

  1. 65

    Stanford HAI updates University newsletter

    Research, policy, education, and event digests from Stanford's Institute for Human-Centered AI.

  2. 66

    The Batch Industry newsletter

    A frequent digest of AI research, products, and industry developments from DeepLearning.AI.

  3. 67

    One Useful Thing Independent commentary

    Ethan Mollick's practical interpretation of AI research, work, education, and experimentation.

  4. 68

    Import AI Industry commentary

    Jack Clark's long-running newsletter on AI research, capability, policy, and security.

Four rules for using this library

  1. Prefer the original source

    If an article says a study proved something, open the study. If a product comparison describes a feature or price, open the vendor's current plan page. Summaries are useful for discovery. Decisions deserve provenance.

  2. Separate capability, adoption, and value

    A benchmark can show that a model can perform a task under test conditions. A usage survey can show that people are trying it. A business report can show reported savings or revenue. Those are three different claims.

  3. Read the boundary, not only the headline

    The most useful sentence in a study may be the one that tells you who was tested, what they did, and what the researchers did not measure. Convert every large claim into a bounded one: “In this population, on this task, with this tool, under these conditions...”

  4. Update your view through practice

    Reading cannot substitute for responsible use. Choose a task you understand, brief it clearly, inspect the return, verify the consequential parts, and run it again. Your own disciplined practice is where general claims become local knowledge.

Start with the Field Guide

Get to Know AI is the durable teaching page: the mental model, where AI is strong, where it fails, how to brief the work, and how to decide whether a result has earned your trust.

Read the field guide → Get the Sunday brief →