{
  "schema_version": 1,
  "reviewed": "2026-11-19",
  "title": "The human journey toward artificial general intelligence",
  "intro": "Can a machine learn its way through a problem its builders never anticipated? Follow the people who have tried to make that question precise, the machines they built, and the tests that changed their minds.",
  "orientation": "Artificial general intelligence (AGI) is a research goal. A working description here is a machine that can learn and use skills across many unfamiliar tasks. Researchers disagree about how wide that range must be, how quickly a system must learn, and how much human help counts. A strong claim names a definition and a test. These records document proposals, experiments, public events and criticisms. Whether any system satisfies a particular definition requires a separate argument.",
  "history_note": "This is a selected history of ideas that matter to the AGI question. Many participants set out to prove a theorem, translate a language, play a game or solve a scientific problem. Their own goals appear in the records. A year alone means the record is dated to that year. Within a year, finer dates are used only when the source supports them.",
  "eras": [
    {
      "id": "ideas",
      "title": "What could a machine do?",
      "span": "1843–1950",
      "bridge": "Before researchers could train a model, they had to describe a procedure, a signal, and a question a machine could be tested against. A mathematical possibility, a model of a neuron and a test of conversation each answer a different question.",
      "question": "If a machine follows a procedure for composing music, what evidence would tell you it learned something?"
    },
    {
      "id": "field",
      "title": "A field takes shape",
      "span": "1955–1973",
      "bridge": "Several ideas competed from the start. One team searched for proofs, another learned from games, and chemists taught a program which structures were plausible. Translators, funders and critics asked what these systems delivered outside a demonstration.",
      "question": "Which part of a system’s apparent skill did a person supply in advance?"
    },
    {
      "id": "methods",
      "title": "Rules, probabilities and learned patterns",
      "span": "1982–1998",
      "bridge": "No single method replaced all the others. A national project pursued logical inference, researchers taught networks to correct their errors, and chess engineers combined search with human expertise. Different designs made different mistakes visible.",
      "question": "What would need to change to move a chess engine, a document reader or a belief network to a new task?"
    },
    {
      "id": "data",
      "title": "Who supplies the data, and who defines success?",
      "span": "2004–2016",
      "bridge": "Field trials in desert driving and shared image collections gave teams common tests. Meanwhile other researchers formalized what they meant by generality. Race organizers, data workers and human opponents all helped set the terms of comparison.",
      "question": "How much of a result belongs to the algorithm, the training examples, the hardware or the person who designed the test?"
    },
    {
      "id": "transfer",
      "title": "One model, many tasks?",
      "span": "2017–2021",
      "bridge": "Language models began reusing what they learned across tasks. At the same time, auditors found gaps hidden by averages, and language communities showed how much it matters who collects data and judges the answers.",
      "question": "When an answer looks right, what shows that the skill transfers to a different person, language or problem?"
    },
    {
      "id": "claims",
      "title": "Assistants, agents and the tests of generality",
      "span": "2022 onward",
      "bridge": "Chat interfaces, feedback from people and access to tools changed what users could do. Evaluators now have to specify the model, the tools it could use, the help it received, the budget and the tasks on which it failed.",
      "question": "What observation would make you revise your judgment of a modern model’s generality?"
    }
  ],
  "routes": [
    {
      "title": "I’m new to AI",
      "text": "Read these eight short entries in order. After each, describe one thing the machine did and one thing a person supplied.",
      "ids": [
        "lovelace-notes",
        "turing-game",
        "samuel-checkers",
        "eliza-dialogue",
        "imagenet-data",
        "transformer-translation",
        "chatgpt-preview",
        "agi-framework"
      ]
    },
    {
      "title": "I want to understand how the machines work",
      "text": "Follow a game position, a wrong prediction and a retrieved passage as three different sources of improvement or information. The final stops connect the history to prompting, retrieval and tool use.",
      "ids": [
        "logic-theorist",
        "perceptron-learning",
        "backprop-learning",
        "temporal-difference",
        "alphago-match",
        "transformer-translation",
        "few-shot-prompting",
        "retrieval-generation",
        "react-tools"
      ]
    },
    {
      "title": "I want to see the people behind the systems",
      "text": "Follow how scientists, government programmes, translators, chess players, crowd workers and human raters shaped the problems and the evidence.",
      "ids": [
        "eniac-programmers",
        "dendral-start",
        "translation-review",
        "japan-fifth",
        "deep-blue-match",
        "imagenet-data",
        "gender-shades-audit",
        "masakhane-translation",
        "instructgpt-raters",
        "bloom-collective"
      ]
    },
    {
      "title": "I want to judge an AGI headline",
      "text": "Compare a formal definition, a surprising game result, a new-task benchmark and a test of sustained work. End by comparing two metrics for the same modern release.",
      "ids": [
        "formal-generality",
        "alphago-match",
        "arc-skill",
        "agi-framework",
        "metr-duration",
        "arc-interactive",
        "index-case"
      ]
    }
  ],
  "lenses": [
    {
      "title": "Can the same agent achieve goals in many settings?",
      "text": "Legg and Hutter define intelligence mathematically by considering rewards across possible environments and giving simpler environments more weight. The ideal measure is a way to reason about generality. Applying it directly to a product is computationally impractical.",
      "sources": [
        "legg-hutter"
      ]
    },
    {
      "title": "How much does the system have to learn from scratch?",
      "text": "Chollet asks how efficiently a system can acquire an unfamiliar skill given its prior knowledge and experience. On his ARC puzzles, someone can change the rule while leaving the colored squares and interface the same. What matters is whether the solver discovers the new rule.",
      "sources": [
        "arc-original"
      ]
    },
    {
      "title": "How good, how broad, and how independent?",
      "text": "Morris and colleagues distinguish level of performance from range of capabilities. They also ask how much autonomy a deployment gives the system. A person can grant software purchasing permission without changing its underlying ability to plan a purchase.",
      "sources": [
        "agi-levels"
      ]
    }
  ],
  "glossary": [
    [
      "Algorithm",
      "A sequence of rules for solving a problem. A chess search can apply rules for legal moves and for choosing which move to examine next."
    ],
    [
      "Heuristic",
      "A shortcut that directs effort toward a promising answer. It can save a search through millions of possibilities and can also miss the best answer."
    ],
    [
      "Model and system",
      "A model is a learned or specified way to make predictions or decisions. A deployed system may add a search engine, documents, instructions, memory or human approval around it."
    ],
    [
      "Training and inference",
      "Training changes a model using examples or feedback. Inference uses the resulting model to produce an answer. Supplying an example in a chat can change the answer without updating the model’s weights."
    ],
    [
      "Weights and parameters",
      "Stored numbers a model uses to process its input. Learning can change the strength of a connection, much as adjusting a mixer changes how strongly a sound contributes to the final mix."
    ],
    [
      "Generalization and transfer",
      "Generalization means doing well on new examples. Transfer asks whether something learned in one setting helps on a different task. Reading a new handwritten digit and learning a new game test different kinds of novelty."
    ],
    [
      "Benchmark and baseline",
      "A benchmark specifies tasks and scoring. A baseline is a reference result, such as the score of a person, an earlier system or a simple rule, measured under stated conditions."
    ],
    [
      "Attention",
      "A calculation that weights relationships between pieces of an input. In translation, the words most relevant to the next output word can change from one step to the next."
    ],
    [
      "Reinforcement learning",
      "Learning to choose actions using feedback called reward. Designers choose what earns a reward, so a high score can reflect their choices as well as the system’s ability."
    ],
    [
      "Agent and autonomy",
      "An agent acts and observes the results. Autonomy describes the freedom people give it to act. Access to more tools can improve an outcome without changing the model."
    ],
    [
      "Training exposure",
      "A test item, or something very close to it, may have appeared during development. An unseen puzzle, a held-out task or an independent replication can make a generality claim easier to evaluate."
    ],
    [
      "Graphics processing unit (GPU)",
      "A processor that can do many similar calculations in parallel. The repeated numerical operations used to train neural networks are a good fit for this kind of hardware."
    ]
  ],
  "activities": [
    {
      "title": "Can a convincing answer come from a simple rule?",
      "text": "On paper, replace “I am …” with “Why are you …?” Try it on “I am curious” and “I am fixing a radio.” What would a listener infer? Compare your two lines with the ELIZA transcript.",
      "answer": "The same substitution works on both strings. The second reply can sound odd. Neither exchange shows whether the rule knows what a radio is.",
      "ids": [
        "eliza-dialogue",
        "turing-game"
      ]
    },
    {
      "title": "Invent a test for transfer",
      "text": "Show a friend two drawings of shapes moved one square to the right. Ask them to transform a new drawing. Then give examples that imply a different rule, such as making a mirror image. Record how many examples or hints each of you needed.",
      "answer": "Report the number of examples, the hidden rule and any hints. Success with a new drawing under the old rule and success after the rule changes are different pieces of evidence.",
      "ids": [
        "arc-skill",
        "arc-interactive"
      ]
    },
    {
      "title": "Check a claim that an agent can work for an hour",
      "text": "Read the METR abstract. Write down whose time was measured, what the stated success probability represents and which kinds of task were sampled. Name one task from your own life that the study did not test.",
      "answer": "Human experts supplied the task durations. The 50% horizon is a modeled success threshold on the sampled tasks. It gives no completion probability for your own untested task.",
      "ids": [
        "metr-duration",
        "index-case"
      ]
    }
  ],
  "questions": [
    {
      "title": "What would count as an unfamiliar task?",
      "text": "Look for a protocol that withholds the rules, documents relevant training exposure and records how many examples or hints the solver receives. Changes to the protocol can alter the conclusion.",
      "ids": [
        "arc-skill",
        "arc-interactive"
      ]
    },
    {
      "title": "How much of a result belongs to the model?",
      "text": "Compare a model with and without access to tools or reference material. Preserve the inputs, budget, model version and scoring method. A gain can come from the surrounding system.",
      "ids": [
        "retrieval-generation",
        "react-tools",
        "index-case"
      ]
    },
    {
      "title": "Can today’s measurements support an arrival date?",
      "text": "A forecast needs a stated definition, assumptions about future progress and a way it could turn out wrong. The timeline records forecasts as attributed proposals only when historically consequential. A milestone requires evidence of something that actually happened.",
      "ids": [
        "formal-generality",
        "agi-framework",
        "metr-duration"
      ]
    }
  ],
  "further": [
    {
      "source": "stanford-history",
      "why": "An accessible independent overview. Compare its compact historical narrative with the original documents linked under the early milestones."
    },
    {
      "source": "agi-volume",
      "why": "An edited collection for readers who want to explore approaches that differ from today’s chat assistants."
    },
    {
      "source": "masakhane",
      "why": "A concrete account of who gets to define a research problem, build the data and judge translation quality."
    },
    {
      "source": "helm",
      "why": "A technical starting point for designing comparisons that report more than one score."
    }
  ],
  "milestones": [
    {
      "id": "lovelace-notes",
      "era": "ideas",
      "date": "1843",
      "kind": "Published notes",
      "title": "Lovelace describes a machine that could work with more than numbers",
      "people": "Ada Lovelace, Charles Babbage and Luigi Menabrea",
      "record": "Babbage designed the Analytical Engine. Lovelace translated Menabrea’s account and added notes explaining how operations could act on relationships expressed as symbols. She worked through a procedure for Bernoulli numbers and considered how musical relationships might be represented.",
      "meaning": "A machine can manipulate a representation of something even when it has no direct experience of that thing. That distinction resurfaces whenever an assistant writes convincingly about the world.",
      "limit": "The Bernoulli table was a written procedure for the proposed engine. The music passage was a conditional possibility. Neither documents a machine composing music in 1843.",
      "sources": [
        "lovelace",
        "lovelace-exhibit"
      ]
    },
    {
      "id": "boole-logic",
      "era": "ideas",
      "date": "1854",
      "kind": "Book publication",
      "title": "Boole gives rules for calculating with logical statements",
      "people": "George Boole",
      "record": "Boole’s Laws of Thought used algebraic signs to express classes and rules for combining them. A human could manipulate a symbol by a stated rule while preserving a logical relationship.",
      "meaning": "Suppose one group contains everything that is red and another everything that is round. The things that are both red and round form another group. Writing that operation symbolically makes part of reasoning into a procedure.",
      "limit": "The book develops a formal system for particular relationships. It leaves open how a person or machine decides which objects, symbols and rules belong in a real problem.",
      "sources": [
        "boole-logic"
      ]
    },
    {
      "id": "universal-machine",
      "era": "ideas",
      "date": "1936-05-28",
      "kind": "Manuscript received",
      "title": "Turing asks what one machine can compute",
      "people": "Alan Turing and the London Mathematical Society",
      "record": "Turing described an abstract machine that follows a finite set of rules. He showed how a universal machine could simulate any machine of the specified kind from its description.",
      "meaning": "A single design can perform many different procedures when supplied with different instructions. This is an important sense of general purpose that predates training machines on data.",
      "limit": "The paper concerns what can be computed in principle. It supplies neither a practical time budget nor a method for learning which procedure solves a new task. The date is the manuscript’s receipt date.",
      "sources": [
        "turing-1936"
      ]
    },
    {
      "id": "logic-neuron",
      "era": "ideas",
      "date": "1943",
      "kind": "Mathematical model",
      "title": "A network of idealized neurons can implement logic",
      "people": "Warren McCulloch and Walter Pitts",
      "record": "McCulloch and Pitts represented nerve cells as simplified units that switch on or off. They described how arrangements of such units could implement logical relationships.",
      "meaning": "A calculation could be spread across a network. Asking how its connections should be learned rather than hand-specified would become a separate research problem.",
      "limit": "The units were deliberately simplified. The construction did not train a network on examples or establish that living brains reason in the same way.",
      "sources": [
        "neuron-1943"
      ]
    },
    {
      "id": "eniac-programmers",
      "era": "ideas",
      "date": "1946-02",
      "kind": "Public demonstration",
      "title": "The ENIAC programmers turn a procedure into a working electronic calculation",
      "people": "Jean Jennings Bartik, Frances Bilas and their fellow ENIAC programmers at the University of Pennsylvania",
      "record": "Six women learned to program ENIAC by working out how to connect its units and set its switches. In her oral history, Bartik recalled getting the machine ready for its public unveiling. The machine’s visible speed rested on extensive preparation and debugging.",
      "meaning": "Here is the labor behind the phrase “the computer calculated.” People had to translate a method into a physical arrangement, find the errors and decide whether to trust its output.",
      "limit": "ENIAC’s public demonstration documented electronic computation. The programmers supplied the procedures. The event is a foundation for later AI work, with a different aim from learning a procedure independently.",
      "sources": [
        "eniac-oral",
        "eniac-penn"
      ]
    },
    {
      "id": "cybernetics-feedback",
      "era": "ideas",
      "date": "1948",
      "kind": "Book publication",
      "title": "Wiener puts feedback at the center of control",
      "people": "Norbert Wiener and the interdisciplinary cybernetics community",
      "record": "Wiener’s Cybernetics brought control and communication in animals and machines into one discussion. In a feedback loop, an action changes what the system senses next.",
      "meaning": "Think of a thermostat. Its last action affects the next temperature reading. A robot or game player must likewise deal with the consequences of its own actions.",
      "limit": "A controller can regulate one variable successfully with a fixed rule. Feedback by itself establishes little about learning a new goal.",
      "sources": [
        "wiener"
      ]
    },
    {
      "id": "turing-game",
      "era": "ideas",
      "date": "1950-10",
      "kind": "Research proposal",
      "title": "Turing makes machine intelligence a question about observable behavior",
      "people": "Alan Turing and the readers of Mind",
      "record": "Turing proposed an imitation game conducted through written exchanges. Later in the same paper he considered constructing a learning machine by educating a simpler starting system.",
      "meaning": "Two distinct questions emerge. Can a person tell the machine’s answers apart in a particular setting? How might the machine acquire the abilities that produced them?",
      "limit": "Turing described a proposed test and a possible research programme. The paper reports no machine passing a universally accepted test for general intelligence.",
      "sources": [
        "turing-1950"
      ]
    },
    {
      "id": "dartmouth-plan",
      "era": "field",
      "date": "1955-08-31",
      "kind": "Dated proposal",
      "title": "Four researchers name an ambitious summer project",
      "people": "John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon",
      "record": "Their proposal used the phrase artificial intelligence and asked for a summer study at Dartmouth the following year. It listed problems including language, neural networks and self-improvement.",
      "meaning": "Several competing approaches already fit under one name. The proposal lets a reader see the researchers’ ambitions in their own document before later successes or disappointments colored the story.",
      "limit": "Their belief that aspects of intelligence could be described precisely enough for a machine was a research premise. The date belongs to the proposal. The meeting followed in 1956.",
      "sources": [
        "dartmouth-proposal"
      ]
    },
    {
      "id": "dartmouth-summer",
      "era": "field",
      "date": "1956",
      "kind": "Summer workshop",
      "title": "Dartmouth gives researchers a shared meeting place",
      "people": "Dartmouth participants and organizers in Hanover, New Hampshire",
      "record": "The summer research project took place in 1956. Dartmouth’s historical account identifies the gathering as an organizing moment for the field.",
      "meaning": "Institutions, funding proposals and a common name helped people working on different questions recognize one another’s work. Earlier work on computation and machine intelligence was already on record.",
      "limit": "The proposal and workshop mark different events. Neither document establishes that all the participants agreed on one method or achieved the proposal’s scientific ambitions.",
      "sources": [
        "dartmouth-meeting",
        "dartmouth-proposal"
      ]
    },
    {
      "id": "logic-theorist",
      "era": "field",
      "date": "1956-10-10",
      "kind": "Technical report",
      "title": "The Logic Theory Machine searches for proofs",
      "people": "Allen Newell, J. C. Shaw and Herbert A. Simon at RAND and Carnegie Tech",
      "record": "The team reported experiments with a program that chose promising steps in a search for proofs in a formal logic. Their report describes the program’s search effort as well as its successes.",
      "meaning": "A machine could produce a sequence of valid steps that a person could check. Choosing which branch to try next mattered because blindly trying everything was expensive.",
      "limit": "The authors supplied a formal language, allowed moves and a target. A correct proof in that setting does not show the machine can formulate a useful question in an unfamiliar subject.",
      "sources": [
        "logic-report"
      ]
    },
    {
      "id": "perceptron-learning",
      "era": "field",
      "date": "1958",
      "kind": "Research paper",
      "title": "The perceptron makes learning part of the design",
      "people": "Frank Rosenblatt at the Cornell Aeronautical Laboratory",
      "record": "Rosenblatt analyzed a model of pattern recognition in which experience changes parts of a network. He asked which environmental distinctions a network could learn to recognize.",
      "meaning": "Instead of writing a separate rule for every example, a designer could choose a learning mechanism and let examples influence the decision. This is a useful way to approach a modern classifier.",
      "limit": "The paper discusses limits on distinguishing relationships. Recognizing a learned pattern and explaining its meaning are different demands on a system.",
      "sources": [
        "perceptron"
      ]
    },
    {
      "id": "samuel-checkers",
      "era": "field",
      "date": "1959",
      "kind": "Research paper",
      "title": "A checkers program improves through play",
      "people": "Arthur Samuel at IBM",
      "record": "Samuel described a program that stored information about positions and experimented with learning how to value a board position. It could use experience from play to guide later move choices.",
      "meaning": "In a game, the rules and eventual outcome are clear. The hard part is recognizing which earlier move deserves credit. That question connects Samuel’s work to later reinforcement learning.",
      "limit": "People still supplied the game, legal moves and a representation of the board. Improvement in checkers leaves transfer to another game untested.",
      "sources": [
        "samuel"
      ]
    },
    {
      "id": "dendral-start",
      "era": "field",
      "date": "1965",
      "kind": "Collaboration begins",
      "title": "Chemists and computer scientists combine their expertise in DENDRAL",
      "people": "Joshua Lederberg, Edward Feigenbaum and collaborators including Bruce Buchanan and Georgia Sutherland",
      "record": "Lederberg dated the start of the Stanford collaboration to early 1965. Later reports described programs that generated candidate molecular structures subject to chemical constraints.",
      "meaning": "A specialist knows which possibilities can be ruled out before a search begins. Making that knowledge explicit can turn a huge search into a tractable scientific aid.",
      "limit": "The entry dates the collaboration. The linked 1968 program record documents a later stage. The success of a chemistry-specific representation leaves other scientific domains open.",
      "sources": [
        "dendral-origin",
        "dendral-program"
      ]
    },
    {
      "id": "eliza-dialogue",
      "era": "field",
      "date": "1966-01",
      "kind": "Research paper",
      "title": "ELIZA makes a conversation feel personal",
      "people": "Joseph Weizenbaum at MIT and the people who conversed with his program",
      "record": "ELIZA matched keywords and rearranged text according to a script. Its well-known DOCTOR script could turn a user’s words into a question, encouraging the user to supply more of the conversation.",
      "meaning": "Read the published dialogue once as a participant and again looking for substitutions. A listener can supply a surprising amount of the meaning attributed to a program.",
      "limit": "The paper documents a conversational effect and the mechanism behind it. Apparent empathy in this example cannot establish understanding or genuine care.",
      "sources": [
        "eliza"
      ]
    },
    {
      "id": "shakey-robot",
      "era": "field",
      "date": "1966",
      "kind": "Research programme",
      "title": "Shakey connects sensing, planning and action",
      "people": "Charles Rosen, Nils Nilsson and the SRI team, supported by ARPA",
      "record": "From 1966 through 1972, SRI researchers built and studied a mobile robot that could perceive parts of a room, plan a route and rearrange simple objects.",
      "meaning": "A robot’s plan has to survive contact with the world. Moving an object changes what the next observation and the next action should be.",
      "limit": "The team used a controlled setting and a limited set of objects and actions. Route planning there did not settle how a robot would handle an unfamiliar home.",
      "sources": [
        "shakey",
        "shakey-funding"
      ],
      "end": "1972"
    },
    {
      "id": "translation-review",
      "era": "field",
      "date": "1966",
      "kind": "Public evaluation",
      "title": "US reviewers challenge the case for automatic translation",
      "people": "The Automatic Language Processing Advisory Committee, translators and US research sponsors",
      "record": "ALPAC examined translation needs, costs and quality in the United States. Its report was skeptical of then-current automatic translation and recommended support for work including aids for human translators and basic research.",
      "meaning": "An institution can change which problems receive attention by asking about the work people actually need done. The translator’s time and the quality of the final text both matter.",
      "limit": "The report concerned a particular US funding and language-service context. Hutchins’s later historical analysis helps distinguish its actual recommendations from the sweeping story that all translation research stopped.",
      "sources": [
        "alpac",
        "alpac-history"
      ]
    },
    {
      "id": "lighthill-review",
      "era": "field",
      "date": "1973",
      "kind": "Commissioned review",
      "title": "The Lighthill report challenges AI’s ability to scale",
      "people": "James Lighthill, the UK Science Research Council and later parliamentary historians",
      "record": "Lighthill observed that a method that works on a small example may run out of time as the number of possibilities grows. The UK review criticized the gap between certain demonstrations and broader ambitions.",
      "meaning": "Imagine trying every route through a few towns, then through fifty. The rule for finding the best route can stay the same while the work required becomes prohibitive.",
      "limit": "A later House of Lords review explicitly treats the causal link between the report and reduced AI research funding as unclear. The document verifies a major criticism. A single worldwide funding-collapse claim needs other evidence.",
      "sources": [
        "lighthill",
        "lighthill-parliament"
      ]
    },
    {
      "id": "japan-fifth",
      "era": "methods",
      "date": "1982",
      "kind": "National research programme",
      "title": "Japan pursues knowledge processing through logic and parallel machines",
      "people": "Japan’s Ministry of International Trade and Industry, ICOT researchers and Shunichi Uchida",
      "record": "Japan launched the Fifth Generation Computer Project. Its researchers explored machines that could apply logical inferences and carry out parts of the work in parallel. Uchida’s 1987 report describes hardware under development.",
      "meaning": "The future of AI once looked like purpose-built machines answering logical queries. This programme shows a serious alternative to today’s training of large neural networks.",
      "limit": "The museum record dates the launch. The technical report establishes what engineers were building. Project ambitions and achieved general intelligence are separate historical claims.",
      "sources": [
        "japan-project",
        "japan-machines"
      ]
    },
    {
      "id": "backprop-learning",
      "era": "methods",
      "date": "1986",
      "kind": "Research paper",
      "title": "Networks learn internal representations from their errors",
      "people": "David Rumelhart, Geoffrey Hinton and Ronald Williams",
      "record": "The authors demonstrated how a difference between a network’s prediction and its target could be propagated backward to adjust connections in several layers. Internal units learned useful patterns in the examples.",
      "meaning": "A builder could provide targets without hand-designing every intermediate feature. This helps explain the change from supplying a whole set of rules to supplying examples and a way to score mistakes.",
      "limit": "The authors demonstrated their procedure on particular problems. Establishing who first developed any of its underlying mathematical ideas would require additional historical evidence.",
      "sources": [
        "backprop"
      ]
    },
    {
      "id": "uncertain-beliefs",
      "era": "methods",
      "date": "1988",
      "kind": "Research monograph",
      "title": "Belief networks make uncertainty part of reasoning",
      "people": "Judea Pearl and the probabilistic reasoning community",
      "record": "Pearl’s book set out how a network of related variables can represent uncertain beliefs and update them when evidence arrives. New observations can change the plausibility of several explanations at once.",
      "meaning": "Suppose a radio goes silent. Discovering that its display is also dark shifts the balance between an audio problem and a power problem. A system can express that change without declaring either diagnosis certain.",
      "limit": "The results depend on how the variables, dependencies and probabilities were specified. A meaningful probability model for one device does not automatically transfer to another.",
      "sources": [
        "pearl",
        "pearl-record"
      ]
    },
    {
      "id": "temporal-difference",
      "era": "methods",
      "date": "1988",
      "kind": "Research paper",
      "title": "A learner updates a prediction before the final outcome arrives",
      "people": "Richard Sutton, building on earlier prediction work including Samuel’s checkers",
      "record": "Sutton analyzed methods that use the difference between successive predictions to update an earlier estimate. His paper explicitly connects this idea to earlier game-playing research.",
      "meaning": "A learner predicting the outcome of a trip can update its estimate when it encounters a delay, before the trip actually ends. This helps explain how feedback can travel across a sequence of decisions.",
      "limit": "The paper proves results under specified conditions and studies prediction problems. A general ability to choose goals or act reliably in arbitrary environments does not follow from those results.",
      "sources": [
        "sutton",
        "samuel"
      ]
    },
    {
      "id": "deep-blue-match",
      "era": "methods",
      "date": "1997-05",
      "kind": "Match result",
      "title": "Deep Blue wins the rematch with world chess champion Garry Kasparov",
      "people": "Garry Kasparov, Feng-hsiung Hsu, Murray Campbell, Joe Hoane and IBM’s human chess advisers",
      "record": "Kasparov won the 1996 match despite losing its opening game. In the 1997 rematch, Deep Blue won the match. Its specialized hardware, search and chess knowledge were the work of people with different skills.",
      "meaning": "A public match made one capability hard to dismiss. The result also gives a clean demonstration of the difference between expert performance in one domain and breadth across domains.",
      "limit": "An individual game, a match and a claim about intelligence are three different units of evidence. The match tested chess under agreed rules.",
      "sources": [
        "deep-blue",
        "deep-blue-museum"
      ]
    },
    {
      "id": "lenet-documents",
      "era": "methods",
      "date": "1998",
      "kind": "Research paper",
      "title": "Neural networks read handwritten documents",
      "people": "Yann LeCun, Léon Bottou, Yoshua Bengio and Patrick Haffner",
      "record": "The authors described convolutional networks for recognizing handwriting and systems that combined learned components for document reading. Looking for similar features at different image positions reduced the need to relearn each feature everywhere.",
      "meaning": "A recognizer can read the same kind of digit written in many different ways. Understanding what a number on a document means in the world requires additional context.",
      "limit": "A system scored on labeled documents establishes performance on that recognition task. It cannot answer arbitrary questions about the author or the transaction from that score alone.",
      "sources": [
        "lenet"
      ]
    },
    {
      "id": "desert-driving",
      "era": "data",
      "date": "2004",
      "kind": "Field challenges",
      "title": "Two desert races turn a failure into a measurable result",
      "people": "DARPA organizers, the competing vehicle teams, and Stanford’s Stanley team led by Sebastian Thrun",
      "record": "No vehicle finished DARPA’s 2004 Grand Challenge. In October 2005 five teams finished a 132-mile course, with Stanford’s Stanley first. The team’s paper describes the combination of sensors, learned methods and probabilistic reasoning used for desert driving.",
      "meaning": "For a learner interested in racing, the finish line makes success concrete. A public failure, a changed course and a second round provide more information than a single successful demonstration.",
      "limit": "The two courses were different. The 2005 finish demonstrated driving under desert race conditions. Everyday traffic, weather and interactions with other drivers require separate evidence.",
      "sources": [
        "grand-challenge",
        "stanley"
      ],
      "end": "2005"
    },
    {
      "id": "agi-collection",
      "era": "data",
      "date": "2007",
      "kind": "Edited research volume",
      "title": "Researchers organize explicitly around artificial general intelligence",
      "people": "Ben Goertzel and Cassio Pennachin, editors, and the volume’s contributors",
      "record": "An edited research volume titled Artificial General Intelligence brought different proposed routes to broadly capable systems together in one place. Its contents preserve approaches beyond the neural networks familiar to today’s readers.",
      "meaning": "A shared goal can organize a research community even when its members disagree about architectures or how to measure success. Read two chapter descriptions and compare the problems they set out to solve.",
      "limit": "The book is evidence that the name and research agenda were in use by 2007. A publication date alone cannot establish who coined a term or prove that any approach worked.",
      "sources": [
        "agi-volume"
      ]
    },
    {
      "id": "formal-generality",
      "era": "data",
      "date": "2007",
      "kind": "Formal definition",
      "title": "Legg and Hutter define intelligence across environments",
      "people": "Shane Legg and Marcus Hutter",
      "record": "Their mathematical definition considers how well an agent achieves goals over a broad collection of possible environments, weighting simpler environments more heavily. It makes an intuitive word into a precise theoretical object.",
      "meaning": "A solver that memorizes the answers to one game can excel there. The question changes when the environment itself can change. This definition makes that difference explicit.",
      "limit": "The idealized measure is not directly computable in full. It is one definition with assumptions about environments, rewards and weighting. Other measurements ask different questions.",
      "sources": [
        "legg-hutter"
      ]
    },
    {
      "id": "imagenet-data",
      "era": "data",
      "date": "2009",
      "kind": "Dataset paper",
      "title": "ImageNet makes image collections and labeling labor visible",
      "people": "Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, Fei-Fei Li and crowd workers",
      "record": "The team organized images using WordNet categories. People working through Amazon Mechanical Turk verified candidate images. The paper reported about 3.2 million images in the collection it described at the time.",
      "meaning": "Training and comparing recognizers at scale depended on choices about words, example images and who checked them. A later audit of ImageNet’s people categories found reasons to remove or reconsider categories and examples.",
      "limit": "A label records a decision within a collection. It can encode ambiguity or harm. The 2009 figure describes that paper’s dataset, not every later ImageNet edition or competition subset.",
      "sources": [
        "imagenet",
        "imagenet-audit"
      ]
    },
    {
      "id": "alexnet-vision",
      "era": "data",
      "date": "2012",
      "kind": "Competition and conference paper",
      "title": "AlexNet shows what data, graphics processors and a deep network can do together",
      "people": "Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton, using the ImageNet community’s data",
      "record": "A deep convolutional network trained on GPUs achieved a striking ImageNet competition result. The team reported top-five classification error, which asks whether the correct category appears among the model’s five highest-ranked guesses.",
      "meaning": "The result brought together earlier ideas about layered recognition, a shared labeled dataset and enough computation to train a large model. Comparing methods on the same task made the improvement legible.",
      "limit": "Picking a category from a supplied list tests recognition under those categories. It does not test whether the machine can describe an unfamiliar scene or learn a new category from one example.",
      "sources": [
        "alexnet",
        "imagenet"
      ]
    },
    {
      "id": "attention-translation",
      "era": "data",
      "date": "2014",
      "kind": "Preprint",
      "title": "A translator learns which source words matter at each step",
      "people": "Dzmitry Bahdanau, Kyunghyun Cho and Yoshua Bengio",
      "record": "The researchers trained a translation model that could weight different parts of a source sentence as it generated each part of the translation. They illustrated these learned alignments in translation examples.",
      "meaning": "Consider two languages that put the verb in different positions. A changing set of weights lets the decoder draw on the relevant source words without requiring a human to pair every word in advance.",
      "limit": "This attention mechanism preceded the 2017 Transformer. The reported experiments concern translation and do not establish that the weights fully explain every decision.",
      "sources": [
        "alignment"
      ]
    },
    {
      "id": "dqn-games",
      "era": "data",
      "date": "2015",
      "kind": "Research experiment",
      "title": "One learning recipe is tested across 49 Atari games",
      "people": "Volodymyr Mnih and colleagues at DeepMind",
      "record": "The team used pixels, game scores and a common learning recipe across 49 Atari games. They trained agents for the individual games and reported scores against human and algorithmic baselines.",
      "meaning": "This broadened the question from whether a specialist could play one game to how broadly its training procedure could be reused. Per-game scores show variation hidden by an average.",
      "limit": "The same recipe across games is different from a single trained player entering every new game with no further practice. The human comparison also depends on the stated playing protocol.",
      "sources": [
        "dqn"
      ]
    },
    {
      "id": "alphago-match",
      "era": "data",
      "date": "2016-03",
      "kind": "Match result",
      "title": "AlphaGo defeats Lee Sedol in a five-game Go match",
      "people": "Lee Sedol, Fan Hui, David Silver and the AlphaGo team, and the expert players whose games supplied examples",
      "record": "AlphaGo combined a search through possible moves with neural networks trained using expert games and self-play. It beat Lee Sedol 4–1 in Seoul. Lee won game four. The earlier Nature paper reported the separate 5–0 match against Fan Hui.",
      "meaning": "The combination matters. Human game records supplied examples, self-play supplied practice and search helped choose a move. One can ask what each contributed and which parts would survive a rule change.",
      "limit": "These results belong to a system trained for Go. The documented original used human game records. Later versions with different training methods require their own evidence.",
      "sources": [
        "alphago-paper",
        "alphago-match"
      ]
    },
    {
      "id": "transformer-translation",
      "era": "transfer",
      "date": "2017",
      "kind": "Conference paper",
      "title": "The Transformer reorganizes how a model handles a sequence",
      "people": "Ashish Vaswani, Noam Shazeer, Niki Parmar and five coauthors",
      "record": "The Transformer used attention to connect positions in a sequence and removed recurrent processing from its core translation architecture. The team evaluated English-to-German and English-to-French translation.",
      "meaning": "Words far apart can affect each other. Processing relationships in parallel during training made this design attractive for large text models. The original translation task is an accessible place to see the idea at work.",
      "limit": "The paper demonstrates a particular architecture and translation results. It did not introduce attention itself or demonstrate a general conversational assistant.",
      "sources": [
        "transformer",
        "alignment"
      ]
    },
    {
      "id": "gender-shades-audit",
      "era": "transfer",
      "date": "2018",
      "kind": "Independent audit",
      "title": "Gender Shades asks whose errors are hidden by an average",
      "people": "Joy Buolamwini, Timnit Gebru and the people represented in the audit dataset",
      "record": "The authors tested three commercial systems that assigned binary gender labels from faces. In their evaluation, error reached 34.7% for the group they labeled darker-skinned female. The highest error for their lighter-skinned male group was 0.8%.",
      "meaning": "Even a headline accuracy that looks strong can hide systematic failures. Reporting the task, the people represented and the subgroup outcomes changes what a reader can reasonably conclude.",
      "limit": "These are the authors’ categories and results for the products and datasets tested. The task was binary label assignment from images. It did not measure identity recognition or establish anyone’s gender.",
      "sources": [
        "gender-shades"
      ]
    },
    {
      "id": "bert-transfer",
      "era": "transfer",
      "date": "2018-10-11",
      "kind": "Preprint",
      "title": "BERT shows how the same pretrained model can be adapted to several language tasks",
      "people": "Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova",
      "record": "BERT learned representations using words on both sides of a position during pretraining. The team then fine-tuned a pretrained model for different tasks. The first preprint appeared in 2018 and the NAACL paper in 2019.",
      "meaning": "Training on ordinary text could provide a reusable starting point. In this setup, adapting the model to a supervised task still involved updating its weights.",
      "limit": "Results came from specified language benchmarks. This training and fine-tuning procedure differs from giving a frozen model examples inside a prompt.",
      "sources": [
        "bert-preprint",
        "bert-conference"
      ]
    },
    {
      "id": "arc-skill",
      "era": "transfer",
      "date": "2019-11-05",
      "kind": "Definition and benchmark",
      "title": "ARC asks how efficiently a solver can learn an unfamiliar rule",
      "people": "François Chollet",
      "record": "Chollet argued for judging the acquisition of skill in light of prior knowledge and practice. His Abstraction and Reasoning Corpus presented small grids where a solver inferred a transformation from a few worked examples.",
      "meaning": "You can try the underlying idea with colored squares and pencil. Solving one new example tests application of a rule. Needing to discover a different rule tests adaptation.",
      "limit": "The benchmark encodes design choices about what prior knowledge people share. A score measures success on this family of tasks under a stated protocol.",
      "sources": [
        "arc-original"
      ]
    },
    {
      "id": "retrieval-generation",
      "era": "transfer",
      "date": "2020-05-22",
      "kind": "Research preprint",
      "title": "Retrieval gives a language model an external source to consult",
      "people": "Patrick Lewis and colleagues",
      "record": "The researchers paired a text generator with retrieval from an external collection. The system selected passages relevant to a question and used them while generating responses on knowledge-intensive tasks.",
      "meaning": "Imagine supplying the correct revision of a lab manual when asking a wiring question. The content the system can retrieve and the model’s ability to interpret it can be evaluated separately.",
      "limit": "Retrieval can return the wrong passage. Even a relevant passage can be misread. A grounded answer requires checking that its actual claim follows from its source.",
      "sources": [
        "rag"
      ]
    },
    {
      "id": "few-shot-prompting",
      "era": "transfer",
      "date": "2020",
      "kind": "Research paper",
      "title": "GPT-3 uses examples inside a prompt",
      "people": "Tom Brown and colleagues at OpenAI and collaborating institutions",
      "record": "GPT-3 was evaluated on tasks presented as text with instructions or a small number of worked examples. In the few-shot setting, the researchers supplied those examples at evaluation time without updating the model’s weights.",
      "meaning": "A reader could now teach the format of a task in the request itself. For an Academy guide, the distinction between an example in the prompt and additional model training matters in everyday use.",
      "limit": "Performance varied across tasks. A few accurate answers can be consistent with several explanations, including a familiar task format. They do not establish reliable learning on every novel problem.",
      "sources": [
        "gpt3"
      ]
    },
    {
      "id": "masakhane-translation",
      "era": "transfer",
      "date": "2020-11",
      "kind": "Community research",
      "title": "African language communities help decide how translation is built and judged",
      "people": "Wilhelmina Nekoto and the Masakhane research community, including translators and participants without formal machine-learning training",
      "record": "The project reported machine-translation benchmarks for more than 30 African languages and human evaluations for about a third of them. Participants helped produce data, models and evaluations.",
      "meaning": "A system cannot be called broadly useful on the strength of a few widely resourced languages. Speakers’ judgments and access to relevant examples change both what can be built and what can be measured.",
      "limit": "The paper reports different levels of evaluation across languages. The existence of a benchmark for a language does not establish that its translations are ready for every setting.",
      "sources": [
        "masakhane"
      ]
    },
    {
      "id": "parrots-debate",
      "era": "transfer",
      "date": "2021",
      "kind": "Research position paper",
      "title": "Researchers challenge the assumption that scale alone delivers understanding",
      "people": "Emily Bender, Timnit Gebru, Angelina McMillan-Major and Shmargaret Shmitchell",
      "record": "The authors argued that expanding language-model training raises questions about data documentation, costs, whose language is represented, and what fluency can lead people to infer. They proposed considering these choices before scaling up.",
      "meaning": "The production of training text and the people affected by a system belong in its technical history. A reader assessing an assistant can ask who supplied the text and evidence used to interpret its output.",
      "limit": "This is a documented scholarly argument. Treat its recommendations as arguments to examine. Individual empirical claims about a later model still require evidence for that model.",
      "sources": [
        "parrots"
      ]
    },
    {
      "id": "alphafold-blind",
      "era": "transfer",
      "date": "2021-07",
      "kind": "Research paper and blind assessment",
      "title": "An independent scientific challenge tests AlphaFold’s predictions",
      "people": "John Jumper and the AlphaFold team, CASP assessors and experimentalists who determined the reference structures",
      "record": "AlphaFold2’s developers reported accurate predictions of the three-dimensional shapes of many proteins in the 2020 CASP14 assessment. The 2021 methods paper and the organizers’ account give two views of the result. The structures used to check the predictions had been withheld from entrants.",
      "meaning": "A scientific test can be designed so that the answer already exists in a lab and is hidden until predictions are submitted. That is a powerful alternative to judging the plausibility of an impressive-looking output.",
      "limit": "This was evidence for structure prediction on the assessed targets. An accurate prediction on this task does not establish the ability to choose and carry out an unrelated scientific investigation.",
      "sources": [
        "alphafold",
        "casp14",
        "casp-validation"
      ]
    },
    {
      "id": "instructgpt-raters",
      "era": "claims",
      "date": "2022",
      "kind": "Research paper",
      "title": "Human demonstrations and rankings teach instruction following",
      "people": "Long Ouyang and colleagues, the human labelers and the people who supplied prompts",
      "record": "For InstructGPT, people wrote examples of desired behavior and ranked outputs. The researchers used this feedback to further train language models. On the tested prompt distribution, raters preferred some outputs from a much smaller tuned model to those from a larger untuned model.",
      "meaning": "What a user finds helpful depends on more than scale or next-word prediction. An evaluation should ask who rated the answers, which prompts they saw and how the alternatives were presented.",
      "limit": "A preference result reflects the selected prompts and raters. An answer can win a preference comparison and still contain a factual error.",
      "sources": [
        "instructgpt"
      ]
    },
    {
      "id": "compute-and-data",
      "era": "claims",
      "date": "2022-03",
      "kind": "Scaling experiment",
      "title": "A smaller model trained on more text can win at the same training budget",
      "people": "Jordan Hoffmann and colleagues at DeepMind",
      "record": "The team compared ways of allocating a fixed amount of computation between model size and training text. Their 70-billion-parameter Chinchilla model was trained on about four times as much data as the 280-billion-parameter Gopher and outperformed it on many reported tasks at a similar training-compute budget.",
      "meaning": "A parameter count alone leaves out how much data the model saw and how resources were spent. Those variables can change the ranking even before considering the tools available to a user.",
      "limit": "The comparison supports a conclusion about the tested budgets, models and tasks. Extending it indefinitely or declaring a general-intelligence threshold would require different evidence.",
      "sources": [
        "chinchilla"
      ]
    },
    {
      "id": "react-tools",
      "era": "claims",
      "date": "2022-10-06",
      "kind": "Research preprint",
      "title": "ReAct interleaves model output, actions and observations",
      "people": "Shunyu Yao and colleagues",
      "record": "ReAct prompted language models to alternate between working through a task, taking permitted actions and using the returned observations. The authors tested settings including question answering with an external knowledge source and interactive simulated tasks.",
      "meaning": "A tool can change what a system knows during an interaction. To compare assistants fairly, record which tools were available, what those tools returned and how many attempts each system had.",
      "limit": "The experiments use specified actions and environments. A displayed reasoning trace is an observable output and cannot by itself establish what caused an internal decision.",
      "sources": [
        "react"
      ]
    },
    {
      "id": "bloom-collective",
      "era": "claims",
      "date": "2022-11-09",
      "kind": "Project research paper",
      "title": "BigScience documents an international approach to building BLOOM",
      "people": "BigScience Workshop participants, multilingual data contributors and the French public computing infrastructure",
      "record": "The BigScience collaboration documented a multilingual model and shared model artifacts and a detailed model card. It trained BLOOM on France’s Jean Zay public supercomputer. The date marks the project paper’s first preprint.",
      "meaning": "Readers can inspect model documentation to ask whose languages were included, how computing time was supplied and what they are permitted to do with the weights. Building a frontier-scale model also required public infrastructure and coordinated labor.",
      "limit": "Access to model weights, training material and the hardware needed to repeat training are distinct. A language appearing in a list is no guarantee of high performance for a particular speaker or task.",
      "sources": [
        "bloom",
        "bloom-card"
      ]
    },
    {
      "id": "helm-evaluation",
      "era": "claims",
      "date": "2022-11-16",
      "kind": "Evaluation study",
      "title": "HELM makes the choice of metrics part of the result",
      "people": "Percy Liang and the Stanford-led HELM collaboration",
      "record": "HELM organized language-model evaluations by scenarios and measured criteria including accuracy, robustness, fairness and efficiency. It reported gaps in what had been evaluated and compared systems under a common framework.",
      "meaning": "A traveler choosing a translation tool might value correctness, speed and cost differently. A single rank silently chooses how much each criterion matters. Several measurements make that choice visible.",
      "limit": "A common framework can improve comparison while still leaving languages, tasks and deployment conditions uncovered. The benchmark’s coverage is part of the finding.",
      "sources": [
        "helm"
      ]
    },
    {
      "id": "chatgpt-preview",
      "era": "claims",
      "date": "2022-11-30",
      "kind": "Public research preview",
      "title": "ChatGPT gives people a conversational way to try instruction following",
      "people": "OpenAI’s ChatGPT team, the trainers and users who provided feedback",
      "record": "OpenAI released ChatGPT as a research preview with dialogue and follow-up turns. Its launch post included examples and acknowledged plausible-sounding errors and sensitivity to wording.",
      "meaning": "A non-programmer could now try the same question with a different instruction and compare the replies. What changed here was public access to an interface as well as the behavior of a trained model.",
      "limit": "This date refers to the ChatGPT preview. A fluent exchange is an observation about the interface and its responses. Checking a particular factual answer requires evidence outside the conversation.",
      "sources": [
        "chatgpt",
        "instructgpt"
      ]
    },
    {
      "id": "multimodal-gpt",
      "era": "claims",
      "date": "2023-03-15",
      "kind": "Technical report",
      "title": "GPT-4’s report describes image and text inputs",
      "people": "OpenAI’s GPT-4 research team",
      "record": "The technical report described a model accepting images and text and producing text, with results across several exams and benchmarks. It also described limitations and withheld substantial training and architecture details.",
      "meaning": "An assistant could in principle answer about a diagram as well as a written paragraph. That creates a second question for a user comparing platforms. Which input types did the tested interface actually expose?",
      "limit": "The paper documents a reported capability. Its publication date cannot establish universal product access, the provenance of every answer, or independent reproduction of its training run.",
      "sources": [
        "gpt4"
      ]
    },
    {
      "id": "agi-framework",
      "era": "claims",
      "date": "2024",
      "kind": "Conference position paper",
      "title": "An AGI framework separates breadth, depth and deployment autonomy",
      "people": "Meredith Ringel Morris and colleagues",
      "record": "A November 2023 preprint appeared as an ICML position paper in 2024. The authors proposed levels of performance and generality and discussed how these interact with the autonomy granted to deployed systems.",
      "meaning": "Try describing a chess champion and a helpful but inconsistent all-purpose assistant on two axes. The chess system can have extraordinary depth and narrow breadth. Giving an assistant more permissions introduces a different consideration.",
      "limit": "The levels are a proposed framework. Placing a system at any level requires a definition of the tasks, comparison group and evidence. Other definitions of AGI remain in use.",
      "sources": [
        "agi-levels",
        "agi-levels-preprint"
      ]
    },
    {
      "id": "deepseek-training",
      "era": "claims",
      "date": "2025-01",
      "kind": "Technical report",
      "title": "DeepSeek distinguishes two recipes for training reasoning models",
      "people": "DeepSeek-AI researchers and the people who prepared data and evaluation tasks",
      "record": "The team reported that R1-Zero applied reinforcement learning to a pretrained base without a preliminary supervised fine-tuning stage. The separate R1 system used initial demonstration data and a multistage training process. The report described strengths and problems such as language mixing in R1-Zero.",
      "meaning": "Small differences in training stages can change what a headline means. Ask which model, which initial training and which task produced the reported result.",
      "limit": "These are the developers’ training descriptions and benchmark results. Omitting one supervised stage does not erase the base model’s training data or the people who constructed the evaluation.",
      "sources": [
        "deepseek"
      ]
    },
    {
      "id": "metr-duration",
      "era": "claims",
      "date": "2025-03-18",
      "kind": "Independent evaluation",
      "title": "METR measures task difficulty in human time",
      "people": "Thomas Kwa and colleagues at METR, and the experts timed on reference tasks",
      "record": "The researchers related agent success on a particular collection of tasks to the time human experts needed for those tasks. Their 50% horizon estimates the human task duration at which the fitted model predicts a one-in-two chance of agent success.",
      "meaning": "If an evaluator says a task lasts an hour, ask whose hour. Human experience, task selection and the required success probability can all change the interpretation.",
      "limit": "The horizon is not the wall-clock runtime of the AI. It is an estimate for the study’s task distribution. Extrapolating a trend to an AGI arrival date adds assumptions the experiment does not test.",
      "sources": [
        "metr"
      ]
    },
    {
      "id": "arc-interactive",
      "era": "claims",
      "date": "2026-03-25",
      "kind": "Benchmark launch",
      "title": "ARC-AGI-3 makes exploration an action-by-action test",
      "people": "ARC Prize’s benchmark team, environment creators and human test participants",
      "record": "ARC Prize launched interactive tasks where a solver must explore unfamiliar environments, infer a goal and act under learned rules. Its subsequent technical report describes how scoring uses human action baselines and accounts for efficiency.",
      "meaning": "A static answer and a sequence of exploratory actions reveal different failures. A solver can now waste moves gathering information, pursue the wrong goal or adapt efficiently after an unexpected observation.",
      "limit": "The launch and the later technical report are separate records. The evidence comes from the benchmark’s creators and describes performance under their tasks and scoring. It supplies no universal threshold for AGI.",
      "sources": [
        "arc3-launch",
        "arc3-study"
      ]
    },
    {
      "id": "index-case",
      "era": "claims",
      "date": "2026-11-17",
      "kind": "Contemporary comparison",
      "title": "Two indices tell different stories about the same release",
      "people": "OpenAI’s GPT-6.2 team and the independent evaluators at Artificial Analysis",
      "record": "After the release of GPT-6.2 Astra and Sol, Artificial Analysis reported that Astra gained five points over GPT-6 Astra on its Intelligence Index. On its Coding Agent Index, Astra was flat against the same predecessor in the Codex harness at maximum effort. The date is the evaluator’s report.",
      "meaning": "This is a worked example for reading a weekly review. Ask which tasks changed, which stayed flat, what tools were allowed and whether the comparison resembles the job you care about.",
      "limit": "These are two named indices at one date with particular evaluation settings. Neither score determines the outcome for all programming tasks or answers whether a model meets an AGI definition.",
      "sources": [
        "aa-nov26",
        "openai-nov26"
      ],
      "weekly": "2026-11-19"
    }
  ],
  "sources": {
    "lovelace": {
      "title": "Notes on the Analytical Engine",
      "credit": "Ada Lovelace, in her translation of Luigi Menabrea",
      "type": "Contemporary publication",
      "url": "https://www.yorku.ca/pclassic/Lovelace/lovelace.htm",
      "look": "Read Note A on representing relations and Note G on Bernoulli numbers.",
      "year": 1843,
      "checked": "2026-11-19"
    },
    "lovelace-exhibit": {
      "title": "Notes",
      "credit": "University of Oklahoma Libraries, Galileo’s World",
      "type": "Museum collection",
      "url": "https://galileo.ou.edu/exhibits/notes.html",
      "look": "See the attribution and the description of the original printed work.",
      "checked": "2026-11-19"
    },
    "boole-logic": {
      "title": "An Investigation of the Laws of Thought, chapter II",
      "credit": "George Boole, Cambridge reproduction",
      "type": "Original publication",
      "url": "https://www.cambridge.org/core/books/abs/an-investigation-of-the-laws-of-thought/signs-and-their-laws/D2FDE9D44CE7E9E589B59106833F15AA",
      "look": "Read the chapter summary on classes, signs and the rules for combining them. The book metadata gives the original year.",
      "year": 1854,
      "checked": "2026-11-19"
    },
    "turing-1936": {
      "title": "On Computable Numbers, with an Application to the Entscheidungsproblem",
      "credit": "Alan Turing",
      "type": "Original paper",
      "url": "https://cs.stanford.edu/people/trevisan/cs172-07/turing36.pdf",
      "look": "See the received and read dates on page 230 and section 6 on the universal machine.",
      "year": 1936,
      "checked": "2026-11-19"
    },
    "neuron-1943": {
      "title": "A Logical Calculus of the Ideas Immanent in Nervous Activity",
      "credit": "Warren McCulloch and Walter Pitts",
      "type": "Original paper",
      "url": "https://www.cs.cmu.edu/~cdm/resources/McCullochPitts1943.pdf",
      "look": "Read the introductory assumptions about all-or-none activity and the relation to logic.",
      "year": 1943,
      "checked": "2026-11-19"
    },
    "eniac-oral": {
      "title": "Oral History of Jean Bartik",
      "credit": "Jean Jennings Bartik interviewed by Gardner Hendrie, Computer History Museum",
      "type": "Participant oral history",
      "url": "https://archive.computerhistory.org/resources/text/Oral_History/Bartik_Jean/102658322.05.01.acc.pdf",
      "look": "Find the discussion of six women developing ENIAC programming and the account of the February public announcement.",
      "year": 2008,
      "checked": "2026-11-19"
    },
    "eniac-penn": {
      "title": "For the Record: Totally Wired",
      "credit": "University of Pennsylvania",
      "type": "Institutional history",
      "url": "https://penntoday.upenn.edu/2006-03-16/record/record-totally-wired",
      "look": "Read the description of the six women who connected units with cables and set switches.",
      "year": 2006,
      "checked": "2026-11-19"
    },
    "wiener": {
      "title": "Cybernetics, or, Control and Communication in the Animal and the Machine",
      "credit": "MIT Libraries on Norbert Wiener’s 1948 book",
      "type": "Library collection",
      "url": "https://libraries.mit.edu/150books/2011/04/04/1948/",
      "look": "Read the account of feedback and control alongside the original book’s publication details.",
      "year": 2011,
      "checked": "2026-11-19"
    },
    "turing-1950": {
      "title": "Computing Machinery and Intelligence",
      "credit": "Alan Turing",
      "type": "Original paper",
      "url": "https://www.cs.mcgill.ca/~dprecup/courses/AI/Materials/turing1950.pdf",
      "look": "Compare section 1 on the imitation game with section 7 on learning machines.",
      "year": 1950,
      "checked": "2026-11-19"
    },
    "dartmouth-proposal": {
      "title": "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence",
      "credit": "John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon",
      "type": "Original proposal",
      "url": "https://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html",
      "look": "Read the dateline and first two paragraphs, then the list of research problems.",
      "year": 1955,
      "checked": "2026-11-19"
    },
    "dartmouth-meeting": {
      "title": "The Founding and Future of AI",
      "credit": "Dartmouth College",
      "type": "Institutional history",
      "url": "https://ai.dartmouth.edu/our-story",
      "look": "See the account of the summer 1956 gathering.",
      "checked": "2026-11-19"
    },
    "logic-report": {
      "title": "Empirical Exploration of the Logic Theory Machine",
      "credit": "Allen Newell, J. C. Shaw and Herbert A. Simon, RAND",
      "type": "Original technical report",
      "url": "https://iiif.library.cmu.edu/file/Newell_box00004_fld00294_doc0001/Newell_box00004_fld00294_doc0001.pdf",
      "look": "The cover of P-951-A is dated October 10, 1956. Read the introduction and the proof-search tests.",
      "year": 1956,
      "checked": "2026-11-19"
    },
    "perceptron": {
      "title": "The Perceptron",
      "credit": "Frank Rosenblatt",
      "type": "Original paper",
      "url": "https://web.stanford.edu/class/psych209a/ReadingsByDate/01_30/Rosenblatt58Perceptron.pdf",
      "look": "Read the learning procedure and the discussion of the model’s limitations.",
      "year": 1958,
      "checked": "2026-11-19"
    },
    "samuel": {
      "title": "Some Studies in Machine Learning Using the Game of Checkers",
      "credit": "Arthur Samuel, IBM",
      "type": "Original paper",
      "url": "https://zoo.cs.yale.edu/classes/cs458/lectures/Samuels.pdf",
      "look": "Compare the sections on rote learning and learning by generalization.",
      "year": 1959,
      "checked": "2026-11-19"
    },
    "dendral-origin": {
      "title": "How DENDRAL Was Conceived and Born",
      "credit": "Joshua Lederberg, Computer History Museum archive",
      "type": "Participant retrospective",
      "url": "https://archive.computerhistory.org/resources/access/text/2024/06/102739347-05-0003-acc.pdf",
      "look": "Find “early 1965” in Lederberg’s chapter on the beginning of the collaboration.",
      "checked": "2026-11-19"
    },
    "dendral-program": {
      "title": "Heuristic DENDRAL",
      "credit": "NASA Technical Reports Server, 1968 conference record",
      "type": "Contemporary technical record",
      "url": "https://ntrs.nasa.gov/citations/19680054644",
      "look": "See the program description, authors and conference date.",
      "year": 1968,
      "checked": "2026-11-19"
    },
    "eliza": {
      "title": "ELIZA",
      "credit": "Joseph Weizenbaum, Communications of the ACM",
      "type": "Original paper",
      "url": "https://courses.cs.umbc.edu/undergraduate/331/papers/eliza.january1966.html",
      "look": "Read the opening dialogue, the description of scripts and the discussion of the user’s interpretation.",
      "year": 1966,
      "checked": "2026-11-19"
    },
    "shakey": {
      "title": "Shakey the Robot",
      "credit": "Nils Nilsson, editor, SRI",
      "type": "Project retrospective",
      "url": "https://www.sri.com/publication/artificial-intelligence-pubs/shakey-the-robot-pub/",
      "look": "Read the opening synopsis for the project dates and the robot’s planning and manipulation tasks.",
      "year": 1984,
      "checked": "2026-11-19"
    },
    "shakey-funding": {
      "title": "Shakey the Robot",
      "credit": "DARPA",
      "type": "Institutional history",
      "url": "https://www.darpa.mil/node/2097",
      "look": "Find Charles Rosen’s proposal and ARPA’s support in 1966.",
      "checked": "2026-11-19"
    },
    "alpac": {
      "title": "Language and Machines",
      "credit": "Automatic Language Processing Advisory Committee, National Research Council",
      "type": "Original public report",
      "url": "https://www.nationalacademies.org/publications/9547",
      "look": "Open the 1966 volume and read its conclusions and recommendations on translation and linguistics.",
      "year": 1966,
      "checked": "2026-11-19"
    },
    "alpac-history": {
      "title": "ALPAC: The (In)famous Report",
      "credit": "John Hutchins, MT News International, 1996",
      "type": "Historical analysis",
      "url": "https://aclanthology.org/www.mt-archive.info/90/MTNI-1996-Hutchins.pdf",
      "look": "Read the opening discussion of the US setting and the section revisiting the recommendations.",
      "year": 1996,
      "checked": "2026-11-19"
    },
    "lighthill": {
      "title": "Artificial Intelligence: A General Survey",
      "credit": "James Lighthill for the UK Science Research Council",
      "type": "Original commissioned report",
      "url": "https://www.aiai.ed.ac.uk/events/lighthill1973/lighthill.pdf",
      "look": "Find the discussion of combinatorial explosion and the report’s treatment of different branches of AI.",
      "year": 1973,
      "checked": "2026-11-19"
    },
    "lighthill-parliament": {
      "title": "AI in the UK: Ready, Willing and Able?",
      "credit": "House of Lords Select Committee on Artificial Intelligence, 2018",
      "type": "Parliamentary retrospective",
      "url": "https://publications.parliament.uk/pa/ld201719/ldselect/ldai/100/10005.htm",
      "look": "Read the historical discussion in paragraphs 23–24, especially the caveat about research funding.",
      "year": 2018,
      "checked": "2026-11-19"
    },
    "japan-project": {
      "title": "Start of the Fifth Generation Computer Project",
      "credit": "Information Processing Society of Japan Computer Museum",
      "type": "Institutional history",
      "url": "https://museum.ipsj.or.jp/en/computer/other/0002.html",
      "look": "Read the opening account of the 1982 launch and the project’s knowledge-processing aims.",
      "checked": "2026-11-19"
    },
    "japan-machines": {
      "title": "Inference Machines in FGCS Project",
      "credit": "Shunichi Uchida, ICOT, 1987",
      "type": "Original technical report",
      "url": "https://www.airc.aist.go.jp/aitec-icot/ICOT/Museum/TRTM/TR/TRpdf/tr0278.pdf",
      "look": "Read the abstract on sequential and parallel inference machines.",
      "year": 1987,
      "checked": "2026-11-19"
    },
    "backprop": {
      "title": "Learning Representations by Back-Propagating Errors",
      "credit": "David Rumelhart, Geoffrey Hinton and Ronald Williams",
      "type": "Original paper",
      "url": "https://sabiod.lis-lab.fr/pub/Hinton_etal_inNature1986_backprop.pdf",
      "look": "Read the opening explanation of error signals and the examples of learned internal representations.",
      "year": 1986,
      "checked": "2026-11-19"
    },
    "pearl": {
      "title": "Probabilistic Reasoning in Intelligent Systems",
      "credit": "Judea Pearl, Morgan Kaufmann",
      "type": "Research monograph",
      "url": "https://shop.elsevier.com/books/probabilistic-reasoning-in-intelligent-systems/pearl/978-0-08-051489-5",
      "look": "See the book description and chapters on belief networks and the propagation of evidence.",
      "checked": "2026-11-19"
    },
    "pearl-record": {
      "title": "Probabilistic Reasoning in Intelligent Systems, 1988 edition",
      "credit": "University of Bologna research bibliography",
      "type": "Bibliographic record",
      "url": "https://apice.unibo.it/bin/view/Publication/ProbabilisticreasoningBook1988",
      "look": "Read the original Morgan Kaufmann edition’s publication year.",
      "year": 1988,
      "checked": "2026-11-19"
    },
    "sutton": {
      "title": "Learning to Predict by the Methods of Temporal Differences",
      "credit": "Richard Sutton",
      "type": "Original paper",
      "url": "https://jmvidal.cse.sc.edu/library/sutton88a.pdf",
      "look": "Read the abstract and introduction on differences between successive predictions and the link to Samuel’s checkers.",
      "year": 1988,
      "checked": "2026-11-19"
    },
    "deep-blue": {
      "title": "Deep Blue",
      "credit": "IBM",
      "type": "Participant institution’s history",
      "url": "https://www.ibm.com/history/deep-blue",
      "look": "Read the separate accounts of the 1996 and 1997 matches and the design team.",
      "checked": "2026-11-19"
    },
    "deep-blue-museum": {
      "title": "Deep Blue team after the 1997 rematch",
      "credit": "Computer History Museum",
      "type": "Museum collection",
      "url": "https://www.computerhistory.org/chess/stl-430b9bbcef379/",
      "look": "Read the match result and the caption identifying the human team.",
      "checked": "2026-11-19"
    },
    "lenet": {
      "title": "Gradient-Based Learning Applied to Document Recognition",
      "credit": "Yann LeCun, Léon Bottou, Yoshua Bengio and Patrick Haffner, Proceedings of the IEEE",
      "type": "Original paper",
      "url": "https://proceedingsoftheieee.ieee.org/gradient-based-learning-applied-to-document-recognition/",
      "look": "Read the description of convolutional networks, handwriting and document systems.",
      "year": 1998,
      "checked": "2026-11-19"
    },
    "grand-challenge": {
      "title": "Grand Challenge",
      "credit": "DARPA",
      "type": "Organizer’s event record",
      "url": "https://www.darpa.mil/about/innovation-timeline/grand-challenge",
      "look": "Compare the March 2004 result with the October 2005 finishers and course.",
      "checked": "2026-11-19"
    },
    "stanley": {
      "title": "Stanley: The Robot That Won the DARPA Grand Challenge",
      "credit": "Sebastian Thrun and the Stanford Racing Team, Journal of Field Robotics",
      "type": "Participant technical paper",
      "url": "https://robots.stanford.edu/papers/thrun.stanley05.pdf",
      "look": "Read the opening description of the two races and the account of the vehicle’s sensors and driving software.",
      "year": 2006,
      "checked": "2026-11-19"
    },
    "agi-volume": {
      "title": "Artificial General Intelligence",
      "credit": "Ben Goertzel and Cassio Pennachin, editors, Springer",
      "type": "Research volume",
      "url": "https://link.springer.com/book/10.1007/978-3-540-68677-4",
      "look": "Read the book overview and compare the different approaches in its contents.",
      "year": 2007,
      "checked": "2026-11-19"
    },
    "legg-hutter": {
      "title": "Universal Intelligence: A Definition of Machine Intelligence",
      "credit": "Shane Legg and Marcus Hutter",
      "type": "Original research paper",
      "url": "https://arxiv.org/abs/0712.3329",
      "look": "Read the definition and the discussion of weighting environments and practical computability.",
      "year": 2007,
      "checked": "2026-11-19"
    },
    "imagenet": {
      "title": "ImageNet: A Large-Scale Hierarchical Image Database",
      "credit": "Jia Deng and colleagues, CVPR",
      "type": "Original dataset paper",
      "url": "https://www.image-net.org/static_files/papers/imagenet_cvpr09.pdf",
      "look": "Read the abstract and data-collection section on WordNet categories and crowd verification.",
      "year": 2009,
      "checked": "2026-11-19"
    },
    "imagenet-audit": {
      "title": "Towards Fairer Datasets",
      "credit": "Kaiyu Yang and colleagues",
      "type": "Authors’ dataset audit",
      "url": "https://arxiv.org/abs/1912.07726",
      "look": "Read the abstract on the people subtree, unsuitable concepts and representation.",
      "year": 2019,
      "checked": "2026-11-19"
    },
    "alexnet": {
      "title": "ImageNet Classification with Deep Convolutional Neural Networks",
      "credit": "Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton, NeurIPS",
      "type": "Original conference paper",
      "url": "https://proceedings.neurips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c-Abstract.html",
      "look": "Read the abstract and the sections on training across GPUs and ImageNet classification.",
      "year": 2012,
      "checked": "2026-11-19"
    },
    "alignment": {
      "title": "Neural Machine Translation by Jointly Learning to Align and Translate",
      "credit": "Dzmitry Bahdanau, Kyunghyun Cho and Yoshua Bengio",
      "type": "Original preprint",
      "url": "https://arxiv.org/abs/1409.0473",
      "look": "Read the abstract and the illustrations of alignment between input words and translated words.",
      "year": 2014,
      "checked": "2026-11-19"
    },
    "dqn": {
      "title": "Human-Level Control through Deep Reinforcement Learning",
      "credit": "Volodymyr Mnih and colleagues, Nature",
      "type": "Original paper",
      "url": "https://deepmind-media.storage.googleapis.com/dqn/DQNNaturePaper.pdf",
      "look": "Read the setup, figure 3 and the per-game results in the extended data.",
      "year": 2015,
      "checked": "2026-11-19"
    },
    "alphago-paper": {
      "title": "Mastering the Game of Go with Deep Neural Networks and Tree Search",
      "credit": "David Silver and colleagues, Nature",
      "type": "Original paper",
      "url": "https://research.google/pubs/mastering-the-game-of-go-with-deep-neural-networks-and-tree-search/",
      "look": "The abstract describes expert games, self-play, search and the match with Fan Hui.",
      "year": 2016,
      "checked": "2026-11-19"
    },
    "alphago-match": {
      "title": "Google DeepMind Challenge Match: Lee Sedol v AlphaGo",
      "credit": "British Go Association",
      "type": "Contemporary match record",
      "url": "https://www.britgo.org/alphago-leesedol",
      "look": "Read the five dated game reports, including Lee Sedol’s game-four win.",
      "checked": "2026-11-19"
    },
    "transformer": {
      "title": "Attention Is All You Need",
      "credit": "Ashish Vaswani and colleagues, NeurIPS",
      "type": "Original conference paper",
      "url": "https://proceedings.neurips.cc/paper_files/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html",
      "look": "Read the abstract and the sections on self-attention and translation trials.",
      "year": 2017,
      "checked": "2026-11-19"
    },
    "gender-shades": {
      "title": "Gender Shades",
      "credit": "Joy Buolamwini and Timnit Gebru, PMLR",
      "type": "Independent evaluation",
      "url": "https://proceedings.mlr.press/v81/buolamwini18a.html",
      "look": "Read the abstract and the results by subgroup for the three commercial systems.",
      "year": 2018,
      "checked": "2026-11-19"
    },
    "bert-preprint": {
      "title": "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding",
      "credit": "Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova",
      "type": "Original preprint",
      "url": "https://arxiv.org/abs/1810.04805",
      "look": "See the October 2018 submission and the description of bidirectional pretraining.",
      "year": 2018,
      "checked": "2026-11-19"
    },
    "bert-conference": {
      "title": "BERT",
      "credit": "Devlin and colleagues, NAACL 2019",
      "type": "Conference publication",
      "url": "https://aclanthology.org/N19-1423/",
      "look": "Use the proceedings date to distinguish the 2019 conference paper from the 2018 preprint.",
      "year": 2019,
      "checked": "2026-11-19"
    },
    "arc-original": {
      "title": "On the Measure of Intelligence",
      "credit": "François Chollet",
      "type": "Research proposal and benchmark",
      "url": "https://arxiv.org/abs/1911.01547",
      "look": "Read the definitions of skill acquisition and priors, then the introduction of the Abstraction and Reasoning Corpus.",
      "year": 2019,
      "checked": "2026-11-19"
    },
    "rag": {
      "title": "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks",
      "credit": "Patrick Lewis and colleagues",
      "type": "Original paper",
      "url": "https://arxiv.org/abs/2005.11401",
      "look": "Read the abstract and model description, including the difference between learned parameters and retrieved documents.",
      "year": 2020,
      "checked": "2026-11-19"
    },
    "gpt3": {
      "title": "Language Models Are Few-Shot Learners",
      "credit": "Tom Brown and colleagues, NeurIPS",
      "type": "Original conference paper",
      "url": "https://proceedings.neurips.cc/paper_files/paper/2020/hash/1457c0d6bfcb4967418bfb8ac142f64a-Abstract.html",
      "look": "Read the abstract and the definition of few-shot demonstrations without gradient updates at test time.",
      "year": 2020,
      "checked": "2026-11-19"
    },
    "masakhane": {
      "title": "Participatory Research for Low-Resourced Machine Translation",
      "credit": "Wilhelmina Nekoto and colleagues, Findings of EMNLP",
      "type": "Original community research",
      "url": "https://aclanthology.org/2020.findings-emnlp.195/",
      "look": "Read the abstract, participant roles and which languages received human evaluation.",
      "year": 2020,
      "checked": "2026-11-19"
    },
    "parrots": {
      "title": "On the Dangers of Stochastic Parrots",
      "credit": "Emily Bender, Timnit Gebru, Angelina McMillan-Major and Shmargaret Shmitchell",
      "type": "Research position paper",
      "url": "https://mlanthology.org/misc/2021/bender2021misc-dangers/",
      "look": "Read the abstract and the recommendations about data documentation, costs and affected communities.",
      "year": 2021,
      "checked": "2026-11-19"
    },
    "alphafold": {
      "title": "Highly Accurate Protein Structure Prediction with AlphaFold",
      "credit": "John Jumper and colleagues, Nature",
      "type": "Original research paper",
      "url": "https://www.nature.com/articles/s41586-021-03819-2",
      "look": "Read the abstract and the account of the blind CASP14 assessment.",
      "year": 2021,
      "checked": "2026-11-19"
    },
    "casp14": {
      "title": "Critical Assessment of Methods of Protein Structure Prediction, Round XIV",
      "credit": "Andriy Kryshtafovych and colleagues, CASP organizers",
      "type": "Independent assessment",
      "url": "https://pubmed.ncbi.nlm.nih.gov/34533838/",
      "look": "Read the abstract for the assessors’ description of CASP14 and the comparison with experimental structures.",
      "year": 2021,
      "checked": "2026-11-19"
    },
    "casp-validation": {
      "title": "How Have AlphaFold2’s Predictions of Protein Structure Been Validated?",
      "credit": "EMBL’s European Bioinformatics Institute",
      "type": "Accessible scientific overview",
      "url": "https://www.ebi.ac.uk/training/online/courses/alphafold/validation-and-impact/how-have-alphafolds-predictions-of-protein-structure-been-validated/",
      "look": "See the explanation of the CASP14 experiment, its comparison with other entrants and the limits of the targets.",
      "checked": "2026-11-19"
    },
    "instructgpt": {
      "title": "Training Language Models to Follow Instructions with Human Feedback",
      "credit": "Long Ouyang and colleagues, NeurIPS",
      "type": "Original conference paper",
      "url": "https://proceedings.neurips.cc/paper_files/paper/2022/hash/b1efde53be364a73914f58805a001731-Abstract.html",
      "look": "Read the labeler demonstration and preference-ranking procedure and the caveats on whose preferences were learned.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "chinchilla": {
      "title": "Training Compute-Optimal Large Language Models",
      "credit": "Jordan Hoffmann and colleagues",
      "type": "Original scaling study",
      "url": "https://arxiv.org/abs/2203.15556",
      "look": "Compare Chinchilla and Gopher under the authors’ matched training-compute budget.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "react": {
      "title": "ReAct: Synergizing Reasoning and Acting in Language Models",
      "credit": "Shunyu Yao and colleagues",
      "type": "Original preprint",
      "url": "https://arxiv.org/abs/2210.03629",
      "look": "Read the action-and-observation examples and the environments used in the experiments.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "bloom": {
      "title": "BLOOM: A 176B-Parameter Open-Access Multilingual Language Model",
      "credit": "BigScience Workshop",
      "type": "Project research paper",
      "url": "https://arxiv.org/abs/2211.05100",
      "look": "Read the abstract and the sections on multilingual data, collaboration and released artifacts.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "bloom-card": {
      "title": "BLOOM model card",
      "credit": "BigScience Workshop",
      "type": "Primary artifact documentation",
      "url": "https://huggingface.co/bigscience/bloom",
      "look": "See the language list, hardware details for Jean Zay and usage terms.",
      "checked": "2026-11-19"
    },
    "helm": {
      "title": "Holistic Evaluation of Language Models",
      "credit": "Percy Liang and colleagues",
      "type": "Original evaluation study",
      "url": "https://arxiv.org/abs/2211.09110",
      "look": "Read the taxonomy of scenarios, metrics and gaps in model coverage.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "chatgpt": {
      "title": "Introducing ChatGPT",
      "credit": "OpenAI",
      "type": "Dated provider announcement",
      "url": "https://openai.com/index/chatgpt/",
      "look": "See the November 30, 2022 dateline, research-preview description and limitations.",
      "year": 2022,
      "checked": "2026-11-19"
    },
    "gpt4": {
      "title": "GPT-4 Technical Report",
      "credit": "OpenAI",
      "type": "Provider technical report",
      "url": "https://arxiv.org/abs/2303.08774",
      "look": "Read the input and output modalities, limitations and the disclosure statement about training and architecture.",
      "year": 2023,
      "checked": "2026-11-19"
    },
    "agi-levels": {
      "title": "Levels of AGI for Operationalizing Progress on the Path to AGI",
      "credit": "Meredith Ringel Morris and colleagues, ICML 2024",
      "type": "Peer-reviewed position paper",
      "url": "https://proceedings.mlr.press/v235/morris24b.html",
      "look": "Read the abstract on depth, breadth and deployment autonomy. Its tables illustrate the proposed framework.",
      "year": 2024,
      "checked": "2026-11-19"
    },
    "agi-levels-preprint": {
      "title": "Levels of AGI for Operationalizing Progress on the Path to AGI",
      "credit": "Morris and colleagues, arXiv",
      "type": "Preprint history",
      "url": "https://arxiv.org/abs/2311.02462",
      "look": "The submission history dates the first preprint to November 2023.",
      "year": 2023,
      "checked": "2026-11-19"
    },
    "deepseek": {
      "title": "DeepSeek-R1",
      "credit": "DeepSeek-AI",
      "type": "Provider technical report",
      "url": "https://arxiv.org/abs/2501.12948",
      "look": "Compare the R1-Zero and R1 training pipelines in the abstract and methods.",
      "year": 2025,
      "checked": "2026-11-19"
    },
    "metr": {
      "title": "Measuring AI Ability to Complete Long Tasks",
      "credit": "Thomas Kwa and colleagues, METR",
      "type": "Independent evaluation",
      "url": "https://arxiv.org/abs/2503.14499",
      "look": "Read the abstract, human time baselines, task selection and uncertainty around success estimates.",
      "year": 2025,
      "checked": "2026-11-19"
    },
    "arc3-launch": {
      "title": "Announcing ARC-AGI-3",
      "credit": "ARC Prize",
      "type": "Benchmark launch",
      "url": "https://arcprize.org/blog/arc-agi-3-launch",
      "look": "See the March 25 launch, interactive format and scoring description.",
      "year": 2026,
      "checked": "2026-11-19"
    },
    "arc3-study": {
      "title": "ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence",
      "credit": "ARC Prize research team",
      "type": "Benchmark authors’ technical report",
      "url": "https://arcprize.org/media/ARC_AGI_3_Technical_Report.pdf",
      "look": "Read the April 22, 2026 report’s abstract and the human baseline and efficiency-scoring sections.",
      "year": 2026,
      "checked": "2026-11-19"
    },
    "aa-nov26": {
      "title": "GPT-6.2 Astra and Sol in the Intelligence Index",
      "credit": "Artificial Analysis",
      "type": "Independent evaluation",
      "url": "https://artificialanalysis.ai/articles/openai-has-the-2-model-in-the-intelligence-index-with-the-release-of-gpt-6-2-astra-and-gpt-6-2-sol",
      "look": "Compare the November 17 Intelligence Index results with the Coding Agent Index under its specified harness and effort.",
      "year": 2026,
      "checked": "2026-11-19"
    },
    "openai-nov26": {
      "title": "Introducing GPT-6.2 Astra and Sol",
      "credit": "OpenAI",
      "type": "Dated provider announcement",
      "url": "https://openai.com/index/introducing-gpt-6-2-astra-sol/",
      "look": "Read the release overview and the descriptions of the two models.",
      "year": 2026,
      "checked": "2026-11-19"
    },
    "stanford-history": {
      "title": "A Short History of AI",
      "credit": "Stanford One Hundred Year Study on Artificial Intelligence, 2016",
      "type": "Independent historical synthesis",
      "url": "https://ai100.stanford.edu/reports/2016-report/appendix-i-short-history-ai",
      "look": "Start with the early decades, then compare the descriptions of expert systems, neural networks and AI winters.",
      "year": 2016,
      "checked": "2026-11-19"
    }
  },
  "corrections": []
}
