# Teach with the AI Atlas Use this pack for a group discussion or independent reading. Start with the map, compare the labels, then check the evidence behind an explanation. Prepared with AI assistance. Not independently fact-checked. Check the sources and qualifications before drawing conclusions. Pack version: 0.22.0. Evidence cutoff: 2026-09-16. [Saved evidence dataset](https://theaiatlas.org/editions/ce5ecc46ea940a3c253e5e3d8271fc3b6dc457033cd25871da6bcaaf75a9b9f3/data.json) This is the dataset cutoff, not the date every source was read. The saved dataset preserves the exact evidence used for this pack. ## Map for discussion The map summarizes selected public statements. Left to right describes the preferred pace of advanced AI development. Bottom to top describes expressed concern about catastrophic harm. A point is our interpretation, not a probability or a membership label. An unplaced person has no dot because the reviewed evidence does not establish both axes. [How to read the map and its limits](https://theaiatlas.org/methodology.md) ## Eight ideas and labels These short explanations include qualifications. A movement, a field of research and a label used by critics are different kinds of thing. ### Artificial intelligence (AI) AI is a field of computing. Its systems use computer programs and hardware to recognize patterns, generate content or make predictions. [Explanatory memorandum on the updated OECD definition of an AI system](https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/03/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system_3c815e51/623da898-en.pdf); [software](https://csrc.nist.gov/glossary/term/software); [Machine Learning Glossary](https://developers.google.com/machine-learning/glossary) Keep in mind: AI and artificial general intelligence (AGI) are different terms. A system can perform a particular task well without having broad human-level abilities. [Levels of AGI for Operationalizing Progress on the Path to AGI](https://arxiv.org/abs/2311.02462) [Read the full explanation and source notes](https://theaiatlas.org/ideas/ai/) ### Language models & large language models (LLMs) An LLM is a model trained to process language. Computer software runs it and, for text generation, produces an answer one piece at a time. [Models](https://huggingface.co/learn/llm-course/en/chapter2/3); [Text generation](https://huggingface.co/docs/transformers/en/llm_tutorial); [LLMs: What's a large language model?](https://developers.google.com/machine-learning/crash-course/llm/transformers) Keep in mind: A model can produce a convincing sentence without checking it against an external source. Search or other tools must be supplied by the surrounding application. [Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf); [Tool use](https://huggingface.co/docs/transformers/en/chat_extras) [Read the full explanation and source notes](https://theaiatlas.org/ideas/llm/) ### Effective accelerationism A movement favoring faster technological growth and opposing centralized restraint. Its claim that acceleration leads to better outcomes is a philosophical position, not a demonstrated safety guarantee. [what the f\* is e/acc](https://effectiveaccelerationism.substack.com/p/what-the-f-is-eacc); [Notes on e/acc principles and tenets](https://effectiveaccelerationism.substack.com/p/repost-notes-on-eacc-principles-and) Keep in mind: Broader techno-optimism does not imply e/acc membership. Its thermodynamic arguments do not establish that AI is safe. Buterin offers a counterpoint: profit alone does not automatically select beneficial directions for technology. [Notes on e/acc principles and tenets](https://effectiveaccelerationism.substack.com/p/repost-notes-on-eacc-principles-and); [My techno-optimism](https://vitalik.eth.limo/general/2023/11/27/techno_optimism.html) [Read the full explanation and source notes](https://theaiatlas.org/ideas/eacc/) ### Deceleration An informal label for wanting slower AI development, often used as criticism. Check which limits a person actually supports. [Notes on e/acc principles and tenets](https://effectiveaccelerationism.substack.com/p/repost-notes-on-eacc-principles-and) Keep in mind: A safety rule, a pause on the most powerful AI and opposition to all technology are different positions. PauseAI says its proposed treaty would usually leave narrow applications unaffected, but also proposes considering longer-term research and hardware restrictions. [PauseAI Proposal (April 2026 version)](https://pauseai.org/proposal) [Read the full explanation and source notes](https://theaiatlas.org/ideas/decel/) ### AI doomer A disputed label for people emphasizing catastrophic AI risks. Concern does not mean believing disaster is inevitable. [Guillaume Verdon: e/acc, AI doomers and effective altruism (transcript \#407)](https://lexfridman.com/guillaume-verdon-transcript/); [Statement on AI Extinction Risk](https://aistatement.com/work/statement-on-ai-extinction-risk) Keep in mind: The CAIS statement calls for reducing extinction risk; it does not say extinction is certain or give a shared probability. The Atlas does not automatically label anyone a doomer. [Statement on AI Extinction Risk](https://aistatement.com/work/statement-on-ai-extinction-risk) [Read the full explanation and source notes](https://theaiatlas.org/ideas/doomer/) ### Effective altruism A community seeking effective ways to help others. Critics question whose measures of benefit count and how much power donors should have. [What is effective altruism?](https://www.effectivealtruism.org/articles/introduction-to-effective-altruism); [Against ‘Effective Altruism’](https://www.radicalphilosophy.com/article/against-effective-altruism); [Response to Effective Altruism](https://www.bostonreview.net/forum/peter-singer-logic-effective-altruism/response-emma-saunders-hastings/) Keep in mind: Alice Crary argues that measures of benefit can miss political causes of harm. Emma Saunders-Hastings warns about donors’ power over people receiving help. EA’s own FAQ replies that institutional change belongs in its scope and that EA need not be utilitarian. These disputes concern how help is defined and governed. EA, longtermism and AI safety are not interchangeable. [Against ‘Effective Altruism’](https://www.radicalphilosophy.com/article/against-effective-altruism); [Response to Effective Altruism](https://www.bostonreview.net/forum/peter-singer-logic-effective-altruism/response-emma-saunders-hastings/); [Frequently asked questions and common objections](https://www.effectivealtruism.org/faqs); [Longtermism](https://www.williammacaskill.com/longtermism) [Read the full explanation and source notes](https://theaiatlas.org/ideas/ea/) ### AI safety & alignment Research aimed at reducing AI harms and improving reliability and alignment with intended goals. A safety goal or framework is not proof that a system is safe. [Concrete Problems in AI Safety](https://arxiv.org/abs/1606.06565); [Risks from Learned Optimization in Advanced Machine Learning Systems](https://arxiv.org/abs/1906.01820); [Strengthening our Frontier Safety Framework](https://deepmind.google/blog/strengthening-our-frontier-safety-framework/) Keep in mind: A safety framework records stated safeguards, not a guarantee of safe implementation or a measured catastrophe probability. [Strengthening our Frontier Safety Framework](https://deepmind.google/blog/strengthening-our-frontier-safety-framework/) [Read the full explanation and source notes](https://theaiatlas.org/ideas/safety/) ### Probability of doom Someone's estimated chance of an AI disaster. Check what outcome, time period and assumptions the number refers to. [Guillaume Verdon: e/acc, AI doomers and effective altruism (transcript \#407)](https://lexfridman.com/guillaume-verdon-transcript/); [Thousands of AI Authors on the Future of AI (version 3)](https://arxiv.org/html/2401.02843v3) Keep in mind: These are judgments, not measured disaster rates. Grace and colleagues asked about different outcomes, loss of control and time limits. They warn that AI expertise does not guarantee forecasting skill and who responds can affect results. They still consider researchers' estimates useful. [Thousands of AI Authors on the Future of AI (version 3)](https://arxiv.org/html/2401.02843v3) [Read the full explanation and source notes](https://theaiatlas.org/ideas/pdoom/) ## Three discussion questions ### 1. What does the map tell you? Choose two people. What does each say about the pace of development, and what does each say about catastrophic risk? - Find a dated source for each axis. Separate a personal statement from an organization’s policy. - Compare the source wording with the map interpretation. What could justify a different placement? Suggested reading: [Yann LeCun](https://theaiatlas.org/people/lecun/); [Yoshua Bengio](https://theaiatlas.org/people/bengio/) ### 2. Who chose the label? Compare e/acc, decel and doomer. Which labels appear in supporters’ descriptions, and which appear in criticism? - Open one cited source for each label. Identify who is speaking and what they are arguing. - Explain what a label leaves out. What evidence would you need before applying it to a person? Suggested reading: [Effective accelerationism](https://theaiatlas.org/ideas/eacc/); [Deceleration](https://theaiatlas.org/ideas/decel/); [AI doomer](https://theaiatlas.org/ideas/doomer/) ### 3. Where could a system go wrong? Imagine an application that drafts email replies. Compare a draft that a person checks with a reply the application sends automatically. - Use the drawing to mark the model, the surrounding software and the decision to send. - Which permissions, checks and records would you want? Explain which question each measure helps answer and what remains uncertain. Suggested reading: [AI system](https://theaiatlas.org/ideas/ai-system/); [Set limits before it takes action](https://theaiatlas.org/evidence.html#practice-access); [Keep a useful record of what happened](https://theaiatlas.org/evidence.html#practice-observe) An example to discuss: drafting an email: Input message → Language model → Draft reply → Person checks → Application sends. Illustrative workflow, not a tested control design. Discuss who should decide whether a message is sent. ## For the discussion leader Ask participants to distinguish the source wording from their own interpretation. Accept a well-supported disagreement with a placement. Leave uncertainty visible and record what additional evidence would help. Finish with one claim, its source, one qualification and one open question. ## Sources for the idea sheet ### Strengthening our Frontier Safety Framework Google DeepMind · First-hand source. Published: 2025-09-22. Updated: 2026-04-17. Material last read: 2026-09-15. https://deepmind.google/blog/strengthening-our-frontier-safety-framework/ The article contains an April 2026 update. A published framework is evidence of stated policy, not an estimate of catastrophe probability. No archive check recorded. ### What is effective altruism? Effective Altruism · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://www.effectivealtruism.org/articles/introduction-to-effective-altruism Definition of a philosophy/community, not a placement on the AI chart. No archive check recorded. ### what the f\* is e/acc e/acc newsletter · First-hand source. Published: 2022-12-26. Material last read: 2026-09-15. https://effectiveaccelerationism.substack.com/p/what-the-f-is-eacc Movement self-description. Participants need not share a single risk estimate. No archive check recorded. ### Notes on e/acc principles and tenets e/acc newsletter / BasedBeffJezos and bayeslord · First-hand source. Published: 2022-10-31. Material last read: 2026-09-16. https://effectiveaccelerationism.substack.com/p/repost-notes-on-eacc-principles-and Date belongs to the repost. Read as a movement's own philosophical claims, not independent scientific validation. Includes adversarial usage of decel. What we read (publisher-page): Principles and tenets, including the embedded July 8, 2022 post using decel as criticism. Read as advocacy, not scientific validation. No archive check recorded. ### Guillaume Verdon: e/acc, AI doomers and effective altruism (transcript \#407) Lex Fridman Podcast · First-hand source. Published: 2023-12-29. Material last read: 2026-09-16. https://lexfridman.com/guillaume-verdon-transcript/ Read the AI doomers and effective altruism sections as examples of contested terminology. An advocate's description of opponents is not a neutral definition of their views. Transcript may contain errors. What we read (publisher-page): Introduction, p(doom) discussion from 57:56, AI doomers section from 2:26:27 and effective altruism section. These are attributed views; the transcript may contain errors. No archive check recorded. ### My techno-optimism Vitalik Buterin · First-hand source. Published: 2023-11-27. Material last read: 2026-09-15. https://vitalik.eth.limo/general/2023/11/27/techno_optimism.html Read via the author's eth.limo site because vitalik.ca did not load. Introduces defensive, differential and decentralization-focused acceleration. Used for ideas, not a new placement of the author. No archive check recorded. ### Longtermism William MacAskill · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://www.williammacaskill.com/longtermism An advocate's account of the moral importance of future people. The page is undated; the book's publication date is not used as the page date. No archive check recorded. ### PauseAI Proposal (April 2026 version) PauseAI · First-hand source. Published: 2026-04-05. Material last read: 2026-09-16. https://pauseai.org/proposal Read the April 5th, 2026 version, including its global pause, training approvals and democratic-control conditions. Used for the movement's published proposal and terminology; its predictions are attributed advocacy. What we read (publisher-page): April 2026 proposal: pause and approval conditions, democratic control, effects of a treaty, and proposed longer-term research and hardware restrictions. No archived copy verified · Archive check 2026-09-15 The Wayback availability lookup returned HTTP 429 (rate limited). No capture was verified; this does not show that no archived copy exists. ### Statement on AI Extinction Risk Center for AI Safety · First-hand source. Published: date unavailable. Material last read: 2026-09-16. https://aistatement.com/work/statement-on-ai-extinction-risk Read the statement after following the safe.ai redirect. Establishes advocacy for treating extinction risk as a priority, not certainty of catastrophe or a common numerical probability. No new signatory claims are inferred. What we read (publisher-page): The extinction-risk statement only. This review does not check the signatory list or current roles. No archive check recorded. ### Levels of AGI for Operationalizing Progress on the Path to AGI Meredith Ringel Morris and coauthors / arXiv · First-hand source. Published: 2023-11-04. Updated: 2025-09-24. Material last read: 2026-09-15. https://arxiv.org/abs/2311.02462 Read abstract, Table 1 and autonomy discussion. Proposed distinctions between breadth, performance and autonomy; no current model classification adopted. No archive check recorded. ### Risks from Learned Optimization in Advanced Machine Learning Systems Evan Hubinger and coauthors / arXiv · First-hand source. Published: 2019-06-05. Updated: 2021-12-01. Material last read: 2026-09-15. https://arxiv.org/abs/1906.01820 Read abstract and revision history introducing mesa-optimization and the relation between learned and training objectives. No archive check recorded. ### Concrete Problems in AI Safety Dario Amodei and coauthors / arXiv · First-hand source. Published: 2016-06-21. Updated: 2016-07-25. Material last read: 2026-09-15. https://arxiv.org/abs/1606.06565 Read the abstract's five accident-risk problems including reward hacking and distributional shift. Does not give a general catastrophe probability. No archive check recorded. ### Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile National Institute of Standards and Technology · First-hand source. Published: 2024-07. Material last read: 2026-09-15. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf Read introduction, section 2.2 on confabulation, and selected MEASURE actions 2.3, 2.5, 2.6, 2.7 and 2.9 concerning evaluation evidence, generalization, citations, generated-code review and safeguards. A voluntary risk-management profile; no claim that all 64 pages or every referenced study was reviewed. No archive check recorded. ### Machine Learning Glossary Google for Developers · First-hand source. Published: date unavailable. Updated: 2026-04-10. Material last read: 2026-09-15. https://developers.google.com/machine-learning/glossary Read the artificial intelligence, deep model, context window, inference and chat entries. Used for terminology, not product performance claims. Updated date follows the earlier displayed page date; initial publication is unspecified. The chat entry was reread during the same-day beginner-content review. Also read the generalization and compute entries. No archive check recorded. ### LLMs: What's a large language model? Google for Developers · First-hand source. Published: date unavailable. Updated: 2026-01-02. Material last read: 2026-09-15. https://developers.google.com/machine-learning/crash-course/llm/transformers Read token prediction, encoder-only and decoder-only variants, and self-attention. Used for architecture and terminology; broad performance comparisons and claims about all LLMs on the teaching page are not adopted. No archive check recorded. ### Models Hugging Face LLM Course · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://huggingface.co/learn/llm-course/en/chapter2/3 Read architecture, weights, checkpoints, loading and saving. Used to distinguish a model's structure and learned values from the application around it. Example code was read, not executed; live page publication date unspecified. No archive check recorded. ### Text generation Hugging Face Transformers documentation · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://huggingface.co/docs/transformers/en/llm_tutorial Read next-token generation, generation settings, temperature, sampling and prompt-format sections. Library options illustrate the process; defaults and suggested temperatures are not treated as universal chatbot behavior. Live page publication date unspecified. No archive check recorded. ### Tool use Hugging Face Transformers documentation · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://huggingface.co/docs/transformers/en/chat_extras Read tool descriptions, model-generated call requests, application execution and returning results to the chat. Example functions were not run. Used for the separation between requesting and executing an action; live page publication date unspecified. No archive check recorded. ### Against ‘Effective Altruism’ Alice Crary / Radical Philosophy · First-hand source. Published: 2021. Material last read: 2026-09-15. https://www.radicalphilosophy.com/article/against-effective-altruism Read the institutional and philosophical critique and discussion of EA replies. These are Crary’s arguments, not findings about every participant or charity. Publication precision is the issue year. No archive check recorded. ### Response to Effective Altruism Emma Saunders-Hastings / Boston Review · First-hand source. Published: 2015-07-01. Material last read: 2026-09-15. https://www.bostonreview.net/forum/peter-singer-logic-effective-altruism/response-emma-saunders-hastings/ Read the complete response on donor power and recipient choice. Historical charity recommendations are not repeated as current advice. The argument warns of a governance problem rather than rejecting every donation. No archive check recorded. ### Frequently asked questions and common objections EffectiveAltruism.org · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://www.effectivealtruism.org/faqs Read the definitions and replies on utilitarianism, uncertainty, systemic change and competing priorities. This is the movement website’s account of its principles, not an independent evaluation of its institutions. No archive check recorded. ### software National Institute of Standards and Technology · First-hand source. Published: date unavailable. Material last read: 2026-09-15. https://csrc.nist.gov/glossary/term/software Read the definitions of software as programs and associated data, and the distinction from physical hardware. The glossary aggregates several contextual definitions, including entries for related hardware and firmware terms; it is not a single universal definition. Referenced standards were not read in full. Live page publication date unspecified. No archive check recorded. ### Explanatory memorandum on the updated OECD definition of an AI system Organisation for Economic Co-operation and Development · First-hand source. Published: 2024-03-05. Material last read: 2026-09-15. https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/03/explanatory-memorandum-on-the-updated-oecd-definition-of-an-ai-system_3c815e51/623da898-en.pdf Read the updated definition and explanatory sections on techniques, human roles, autonomy, physical and virtual environments, inputs, models and outputs (pages 4 and 6 to 9). Publication date follows the OECD publication landing page. Used to distinguish models from systems and learned methods from knowledge-based approaches, not as a claim about human-like understanding or a legal classification of a particular product. No archive check recorded. ### Thousands of AI Authors on the Future of AI (version 3) Katja Grace and colleagues / arXiv · First-hand source. Published: 2024-01-05. Updated: 2025-10-08. Material last read: 2026-09-16. https://arxiv.org/html/2401.02843v3 Read selected sections only. Survey estimates express respondents' beliefs; they do not measure future disaster rates. What we read (publisher-page): Abstract, introduction, section 2, sections 4.2–4.3 and 5.2 of version 3; not the full paper or appendices. No archive check recorded. Check the latest explanations and published changes before reusing an older pack. [Content changes](https://theaiatlas.org/changes/).