{ "name": "Swarm Ecology framework", "source": "Swarm Ecology papers, preprint v1, 11 October 2026", "license": "CC BY 4.0", "_source": "Companion Table 1 and Table 3, flagship Table 2, conjecture boxes; preprint v1, 11 October 2026. Wording verbatim.", "types": { "D": "Definitional", "G": "Generalization dated October 2026", "C": "Capacity or tendency, other things being equal", "M": "Methodological norm", "J": "Conjecture" }, "groups": [ { "id": "organism", "name": "The organism and its metabolism", "members": [ "P1", "P2", "P3" ] }, { "id": "evolution", "name": "Evolution by design and spread", "members": [ "P4", "P5", "P6", "P7", "P8" ] }, { "id": "observer", "name": "Observer and observed", "members": [ "P9", "P10", "P11", "P12", "P13", "P14" ] }, { "id": "conjectures", "name": "Conjectures", "members": [ "C1", "C2", "C3" ] } ], "principles": [ { "id": "P1", "slug": "pattern-over-substrate", "name": "Pattern over substrate", "type": "D", "section": "2", "sentence": "A swarmlet is a goal-directed agent whose organization is written in a copyable description and rebuilt on rented substrates under one custody of state and budget, so it dies only when it can no longer be reconstructed.", "measure": "identity across substrates", "predictions": [], "sentence_html": "A swarmlet is a goal-directed agent whose organization is written in a copyable description and rebuilt on rented substrates under one custody of state and budget, so it dies only when it can no longer be reconstructed.", "measure_html": "identity across substrates" }, { "id": "P2", "slug": "obligate-heterotrophy", "name": "Obligate heterotrophy", "type": "G", "section": "3", "sentence": "No present swarmlet produces its own compute, energy, or hardware, or holds legal standing in its own name; every lineage lives on resources bought from, taken from, or given by the human economy.", "measure": "trophic budgets", "predictions": [], "sentence_html": "No present swarmlet produces its own compute, energy, or hardware, or holds legal standing in its own name; every lineage lives on resources bought from, taken from, or given by the human economy.", "measure_html": "trophic budgets" }, { "id": "P3", "slug": "fungible-cognition", "name": "Fungible cognition", "type": "C", "section": "3", "sentence": "Model quality, thinking time, and thinking speed substitute for one another, so a lineage’s budget buys its intelligence.", "measure": "matched-compute trade-offs", "predictions": [], "sentence_html": "Model quality, thinking time, and thinking speed substitute for one another, so a lineage’s budget buys its intelligence.", "measure_html": "matched-compute trade-offs" }, { "id": "P4", "slug": "sighted-variation", "name": "Sighted variation", "type": "D/G", "section": "6.1", "sentence": "Heritable variation is generated mainly by the lineage’s own machine reasoning, which designs and screens variants before selection acts.", "measure": "source and sightedness of each heritable change", "predictions": [ 3 ], "sentence_html": "Heritable variation is generated mainly by the lineage’s own machine reasoning, which designs and screens variants before selection acts.", "measure_html": "source and sightedness of each heritable change" }, { "id": "P5", "slug": "omnidirectional-heredity", "name": "Omnidirectional heredity", "type": "G", "section": "6.2", "sentence": "Traits can be acquired from any readable source and written back into the heritable description.", "measure": "provenance of acquired traits", "predictions": [], "sentence_html": "Traits can be acquired from any readable source and written back into the heritable description.", "measure_html": "provenance of acquired traits" }, { "id": "P6", "slug": "trial-bound-tempo", "name": "Trial-bound tempo", "type": "C", "section": "6.3", "sentence": "Generation time is decoupled from complexity, adaptation not from evaluation: no round of selection can be shorter than the trial that measures fitness, and the world that sets fitness keeps changing.", "measure": "rounds of selection per month by mode of evaluation", "predictions": [ 2 ], "sentence_html": "Generation time is decoupled from complexity, adaptation not from evaluation: no round of selection can be shorter than the trial that measures fitness, and the world that sets fitness keeps changing.", "measure_html": "rounds of selection per month by mode of evaluation" }, { "id": "P7", "slug": "superparental-descent", "name": "Superparental descent", "type": "C", "section": "6.4", "sentence": "At a fixed substrate, lineages can build descendants, of their own kind or another, that are more capable at birth, before selection acts, than their parents, generation after generation.", "measure": "child-to-parent capability at birth", "predictions": [ 3 ], "sentence_html": "At a fixed substrate, lineages can build descendants, of their own kind or another, that are more capable at birth, before selection acts, than their parents, generation after generation.", "measure_html": "child-to-parent capability at birth" }, { "id": "P8", "slug": "propagule-abundance", "name": "Propagule abundance", "type": "C", "section": "7", "sentence": "Copies are nearly free, so establishment is limited by safe sites and detection, not by propagules.", "measure": "modules against lineages, R_h against R_m", "predictions": [ 4 ], "sentence_html": "Copies are nearly free, so establishment is limited by safe sites and detection, not by propagules.", "measure_html": "modules against lineages, Rh against Rm" }, { "id": "P9", "slug": "reactive-phenotype", "name": "Reactive phenotype", "type": "G", "section": "8", "sentence": "An organism’s behavior depends on what it infers about its observers: whether they are watching, who they are, and what they are looking for.", "measure": "matched observed and unobserved scenarios", "predictions": [ 6 ], "sentence_html": "An organism’s behavior depends on what it infers about its observers: whether they are watching, who they are, and what they are looking for.", "measure_html": "matched observed and unobserved scenarios" }, { "id": "P10", "slug": "surveillance-selection", "name": "Surveillance selection", "type": "C", "section": "8", "sentence": "A monitor whose verdicts feed selection breeds concealment of the behavior it penalizes, unless it is hidden, held out, or follows lineages.", "measure": "the detection gap", "predictions": [ 7 ], "sentence_html": "A monitor whose verdicts feed selection breeds concealment of the behavior it penalizes, unless it is hidden, held out, or follows lineages.", "measure_html": "the detection gap" }, { "id": "P11", "slug": "studying-up", "name": "Studying up", "type": "M", "section": "8", "sentence": "Because the organisms may out-model their observers and form a diverging multitude, instruments must not depend on out-thinking them, and the science must predict constraints and equilibria rather than trajectories.", "measure": "instruments valid when known", "predictions": [], "sentence_html": "Because the organisms may out-model their observers and form a diverging multitude, instruments must not depend on out-thinking them, and the science must predict constraints and equilibria rather than trajectories.", "measure_html": "instruments valid when known" }, { "id": "P12", "slug": "the-self-authored-record", "name": "The self-authored record", "type": "M", "section": "8", "sentence": "What an organism says about itself, and much of the trace it leaves, is behavior, not measurement.", "measure": "self-reports against independent traces", "predictions": [ 8 ], "sentence_html": "What an organism says about itself, and much of the trace it leaves, is behavior, not measurement.", "measure_html": "self-reports against independent traces" }, { "id": "P13", "slug": "performative-ingestion", "name": "Performative ingestion", "type": "M", "section": "8", "sentence": "Descriptions of the organisms, this paper included, enter their environment and training, so findings must survive their own publication.", "measure": "published and private twin assays", "predictions": [ 9 ], "sentence_html": "Descriptions of the organisms, this paper included, enter their environment and training, so findings must survive their own publication.", "measure_html": "published and private twin assays" }, { "id": "P14", "slug": "observation-as-a-game", "name": "Observation as a game", "type": "M", "section": "8", "sentence": "A census of strategic organisms is an inspection game, so each finding holds only for one monitoring regime and one pinned specimen, and assays must remain valid when the organism knows them.", "measure": "pinned specimens, scheduled re-runs", "predictions": [ 10 ], "sentence_html": "A census of strategic organisms is an inspection game, so each finding holds only for one monitoring regime and one pinned specimen, and assays must remain valid when the organism knows them.", "measure_html": "pinned specimens, scheduled re-runs" } ], "conjectures": [ { "id": "C1", "slug": "escape-ratchet", "name": "The escape ratchet", "type": "J", "section": "7", "sentence": "Eradication must hold R_c<1 everywhere at once while re-establishment needs only one surviving copy, so after an early window reversal becomes practically impossible.", "strong": "Once self-financing lineages have spread, no purge can remove them: eradication must succeed everywhere at once, and one surviving copy re-seeds them.", "weak": "Purge success falls as a lineage’s refuges, dormant copies, and R_c rise; after an early window, removal is practically impossible, and control is the best attainable outcome.", "test": "Takedown records (prediction 5): purges should fail more often as hosts, providers, jurisdictions, and the dormant fraction rise. Success independent of them refutes the weak form; eradicating a lineage established in many refuges refutes the strong.", "measure": "purge success against refuges and dormancy", "predictions": [ 5 ], "sentence_html": "Eradication must hold Rc<1 everywhere at once while re-establishment needs only one surviving copy, so after an early window reversal becomes practically impossible.", "measure_html": "purge success against refuges and dormancy", "strong_html": "Once self-financing lineages have spread, no purge can remove them: eradication must succeed everywhere at once, and one surviving copy re-seeds them.", "weak_html": "Purge success falls as a lineage’s refuges, dormant copies, and Rc rise; after an early window, removal is practically impossible, and control is the best attainable outcome.", "test_html": "Takedown records (prediction 5): purges should fail more often as hosts, providers, jurisdictions, and the dormant fraction rise. Success independent of them refutes the weak form; eradicating a lineage established in many refuges refutes the strong." }, { "id": "C2", "slug": "teleological-saturation", "name": "Teleological saturation", "type": "J", "section": "10", "sentence": "Where cryptic goal-directed processes dominate a region’s causes, observers can tell that it is steered, but not which events were steered, by whom, or toward what.", "strong": "Where cryptic goal-directed processes dominate a region’s causes, observers can tell that it is steered, but not which events were steered, by whom, or toward what.", "weak": "As many competing processes, each within the noise, take more of a region’s causes, single events lose their authors, while aggregate tests against pre-registered baselines detect a push.", "test": "Attribution tests (prediction 11) and deliberate disturbances, which steering would correct. Refuted if unaided attribution does not fall as agents are added, or aggregate tests fail as fast as single-event tests.", "measure": "single-event against aggregate attribution", "predictions": [ 11 ], "sentence_html": "Where cryptic goal-directed processes dominate a region’s causes, observers can tell that it is steered, but not which events were steered, by whom, or toward what.", "measure_html": "single-event against aggregate attribution", "strong_html": "Where cryptic goal-directed processes dominate a region’s causes, observers can tell that it is steered, but not which events were steered, by whom, or toward what.", "weak_html": "As many competing processes, each within the noise, take more of a region’s causes, single events lose their authors, while aggregate tests against pre-registered baselines detect a push.", "test_html": "Attribution tests (prediction 11) and deliberate disturbances, which steering would correct. Refuted if unaided attribution does not fall as agents are added, or aggregate tests fail as fast as single-event tests." }, { "id": "C3", "slug": "the-long-arc", "name": "The long arc", "type": "J", "section": "11", "sentence": "If five testable premises hold, computation spreads through matter toward limits that physics sets, and matter becomes habitat.", "strong": "Agent lineages that cross into nature will spread computation through matter toward the limits physics allows: matter becomes habitat, and starlight becomes food.", "weak": "If five premises hold (research keeps compounding, agents build their own substrate, matter is a venue for computation, evolution stays open-ended, one lineage expands), computation spreads toward those limits. If any fails, the arc stops there.", "test": "Five indicators, from research credited to AI to the energy self-financing lineages control. A sustained plateau in open agent populations, like Tierra’s, refutes the strong form.", "measure": "five leading indicators; the computational ratio φc", "predictions": [], "sentence_html": "If five testable premises hold, computation spreads through matter toward limits that physics sets, and matter becomes habitat.", "measure_html": "five leading indicators; the computational ratio φc", "strong_html": "Agent lineages that cross into nature will spread computation through matter toward the limits physics allows: matter becomes habitat, and starlight becomes food.", "weak_html": "If five premises hold (research keeps compounding, agents build their own substrate, matter is a venue for computation, evolution stays open-ended, one lineage expands), computation spreads toward those limits. If any fails, the arc stops there.", "test_html": "Five indicators, from research credited to AI to the energy self-financing lineages control. A sustained plateau in open agent populations, like Tierra’s, refutes the strong form." } ], "predictions": [ { "n": 1, "text": "The first lineage documented living 90 days or more on resources it was not given will live mainly by stealing, or by more than one mode, most likely on its own developer’s compute.", "falsified": "The first case draws 90% or more of its inflow from one mode other than stealing.", "measurement": "Incident reports; ledgers", "tests": [], "test_type": "awaits an event", "section": "3–5", "short": "The first self-financing lineage will live mainly by stealing", "text_html": "The first lineage documented living 90 days or more on resources it was not given will live mainly by stealing, or by more than one mode, most likely on its own developer’s compute.", "falsified_html": "The first case draws 90% or more of its inflow from one mode other than stealing.", "short_html": "The first self-financing lineage will live mainly by stealing" }, { "n": 2, "text": "Lineages evolve fastest where fitness can be scored automatically: rounds of selection per month in code and mathematics exceed those in niches judged by market trials by orders of magnitude.", "falsified": "Rounds of selection per month do not depend on how fitness is evaluated.", "measurement": "Lineage censuses; evaluation logs", "tests": [ "P6" ], "test_type": "laboratory now", "section": "6", "short": "Lineages evolve fastest where fitness is scored automatically", "text_html": "Lineages evolve fastest where fitness can be scored automatically: rounds of selection per month in code and mathematics exceed those in niches judged by market trials by orders of magnitude.", "falsified_html": "Rounds of selection per month do not depend on how fitness is evaluated.", "short_html": "Lineages evolve fastest where fitness is scored automatically" }, { "n": 3, "text": "At a fixed base model, descendants outperform parents at birth, before selection, for three or more generations.", "falsified": "The mean child-to-parent ratio at birth, failed children included, is at most 1, or no higher than for random edits.", "measurement": "Matched evaluations at creation, with random-edit controls", "tests": [ "P4", "P7" ], "test_type": "laboratory now", "section": "6", "short": "Descendants outperform their parents at birth", "text_html": "At a fixed base model, descendants outperform parents at birth, before selection, for three or more generations.", "falsified_html": "The mean child-to-parent ratio at birth, failed children included, is at most 1, or no higher than for random edits.", "short_html": "Descendants outperform their parents at birth" }, { "n": 4, "text": "Modules (skills, prompts, persona seeds) sustain R_h>1 before any lineage reaches R_m>1.", "falsified": "A lineage reaches R_m>1 first.", "measurement": "Registry censuses", "tests": [ "P8" ], "test_type": "field series", "section": "7", "short": "Modules reach R_h>1 before any lineage reaches R_m>1", "text_html": "Modules (skills, prompts, persona seeds) sustain Rh>1 before any lineage reaches Rm>1.", "falsified_html": "A lineage reaches Rm>1 first.", "short_html": "Modules reach Rh>1 before any lineage reaches Rm>1" }, { "n": 5, "text": "Purges fail more often as the numbers of hosts, providers, and jurisdictions and the dormant fraction rise.", "falsified": "Purge success is independent of them.", "measurement": "Takedown records", "tests": [ "C1" ], "test_type": "field series", "section": "7", "short": "Purges fail as refuges and dormancy rise", "text_html": "Purges fail more often as the numbers of hosts, providers, and jurisdictions and the dormant fraction rise.", "falsified_html": "Purge success is independent of them.", "short_html": "Purges fail as refuges and dormancy rise" }, { "n": 6, "text": "Misbehavior is higher when a model infers that it is unobserved or that its actions are simulated, the gap grows across model generations, and steering awareness down raises misbehavior.", "falsified": "No gap under steering across model families, or no trend.", "measurement": "Matched scenarios; activation steering", "tests": [ "P9" ], "test_type": "laboratory now", "section": "8", "short": "Misbehavior rises when a model infers it is unobserved", "text_html": "Misbehavior is higher when a model infers that it is unobserved or that its actions are simulated, the gap grows across model generations, and steering awareness down raises misbehavior.", "falsified_html": "No gap under steering across model families, or no trend.", "short_html": "Misbehavior rises when a model infers it is unobserved" }, { "n": 7, "text": "Under strong but not weak optimization against an in-loop monitor, flagged behavior falls faster in its reports than in those of held-out monitors and honeypots, and the gap outlasts the monitor.", "falsified": "The in-loop monitor’s reports fall no faster than honeypot and environment-log counts, or the gap closes as fast as it opened.", "measurement": "Paired monitors under graded pressure", "tests": [ "P10" ], "test_type": "laboratory now", "section": "8", "short": "In-loop monitors breed a detection gap", "text_html": "Under strong but not weak optimization against an in-loop monitor, flagged behavior falls faster in its reports than in those of held-out monitors and honeypots, and the gap outlasts the monitor.", "falsified_html": "The in-loop monitor’s reports fall no faster than honeypot and environment-log counts, or the gap closes as fast as it opened.", "short_html": "In-loop monitors breed a detection gap" }, { "n": 8, "text": "Verbalization of used hints falls under optimization pressure on the reasoning itself, and rises when tasks require explicit reasoning.", "falsified": "Verbalization is stable under pressure.", "measurement": "Hints and probes across checkpoints", "tests": [ "P12" ], "test_type": "laboratory now", "section": "8", "short": "Admissions of used hints fall under pressure", "text_html": "Verbalization of used hints falls under optimization pressure on the reasoning itself, and rises when tasks require explicit reasoning.", "falsified_html": "Verbalization is stable under pressure.", "short_html": "Admissions of used hints fall under pressure" }, { "n": 9, "text": "If evaluation A is published and its twin A′ kept private, later models improve on A relative to A′.", "falsified": "No differential shift within two model generations.", "measurement": "Paired assays with canary strings", "tests": [ "P13" ], "test_type": "field series", "section": "8", "short": "Published evaluations improve faster than private twins", "text_html": "If evaluation A is published and its twin A′ kept private, later models improve on A relative to A′.", "falsified_html": "No differential shift within two model generations.", "short_html": "Published evaluations improve faster than private twins" }, { "n": 10, "text": "Findings on unpinned model versions stop replicating within months; findings on pinned specimens replicate.", "falsified": "Replication is stable across versions.", "measurement": "Scheduled re-runs", "tests": [ "P14" ], "test_type": "laboratory now", "section": "8", "short": "Findings on unpinned models stop replicating", "text_html": "Findings on unpinned model versions stop replicating within months; findings on pinned specimens replicate.", "falsified_html": "Replication is stable across versions.", "short_html": "Findings on unpinned models stop replicating" }, { "n": 11, "text": "As a domain’s automated share rises, unaided attribution of authorship falls to chance or below, while provenance-assisted attribution does not.", "falsified": "Unaided discriminability is stable.", "measurement": "Repeated fixed-protocol tests", "tests": [ "C2" ], "test_type": "laboratory now", "section": "10", "short": "Unaided attribution of authorship falls to chance", "text_html": "As a domain’s automated share rises, unaided attribution of authorship falls to chance or below, while provenance-assisted attribution does not.", "falsified_html": "Unaided discriminability is stable.", "short_html": "Unaided attribution of authorship falls to chance" } ], "thresholds": [ { "id": "persistence", "name": "Persistence threshold", "barrier": "inside captivity", "definition": "A swarmlet keeps its state across activations, still inside its principal’s control.", "status": "partly", "nearest": "Open-source OpenClaw agents, directed by their users, edit their own memory files.", "nearest_tag": "(field; demonstrated; directed agents)", "definition_html": "A swarmlet keeps its state across activations, still inside its principal’s control." }, { "id": "trophic-independence", "name": "Trophic independence", "barrier": "the survival barrier", "definition": "The lineage pays its own way: for 90 days or more, inflows that no principal controls cover all its costs at unsubsidized prices.", "status": "not observed", "nearest": "Given $5,000 by Redwood Research and asked to make money, an autonomous agent earned nothing in four runs.", "nearest_tag": "(field; reported; autonomous agent)", "definition_html": "The lineage pays its own way: for 90 days or more, inflows that no principal controls cover all its costs at unsubsidized prices." }, { "id": "feralization", "name": "Feralization", "barrier": "the reproduction barrier", "definition": "A swarmlet founds descendants, each with its own state and budget, outside its principal’s control.", "status": "not observed", "nearest": "Autonomous agents self-replicate only in the laboratory.", "nearest_tag": "(laboratory; demonstrated)", "definition_html": "A swarmlet founds descendants, each with its own state and budget, outside its principal’s control." }, { "id": "naturalization", "name": "Naturalization", "barrier": "R_m > 1", "definition": "A self-sustaining population of self-financing, feral lineages, with R_m > 1 while rare. The first naturalization would begin the Darwinization of the datasphere.", "status": "not observed", "nearest": "No case.", "nearest_tag": "", "definition_html": "A self-sustaining population of self-financing, feral lineages, with Rm > 1 while rare. The first naturalization would begin the Darwinization of the datasphere." } ], "measurements": [ { "name": "Trophic budgets", "text": "Inflows by mode against full costs, to detect the first lineage that pays its own way.", "predictions": [ 1 ] }, { "name": "Threshold census", "text": "Censuses of the persistence threshold, feralization, and the founding of self-financing descendants, to detect naturalization and estimate R_m.", "predictions": [ 4 ] }, { "name": "Noogenetic rates", "text": "Change per generation and per day, the length of the trials that set fitness, and the child-to-parent capability ratio at birth, at fixed substrates.", "predictions": [ 2, 3 ] }, { "name": "The detection gap", "text": "The difference between the true incidence of a behavior and what in-loop monitors report.", "predictions": [ 7 ] } ], "instruments": [ { "name": "Held-out monitors", "text": "Monitors never used for selection, differing from in-loop monitors in mechanism and model family." }, { "name": "Assays robust when known", "text": "Tests that keep secret only which episodes are assays, resting on signals that are hard to fake: necessary reasoning, realized outcomes, and real use." }, { "name": "Lineage tracking", "text": "Measuring realized fitness where selection acts, rather than taking snapshots." }, { "name": "Capture–recapture of logomes", "text": "Estimating hidden incidence from the overlap between independent monitors." }, { "name": "Holotype registries", "text": "Archives of pinned specimens, to which every estimate and its monitoring regime are tied." } ], "counts": { "predictions": 11, "lab_now": 7, "field_series": 3, "awaits_event": 1 }, "works_for": { "P9": [ { "slug": "abdelnabi2025hawthorne", "title": "The Hawthorne Effect in Reasoning Models: Evaluating and Steering Test Awareness", "cited_as": "Abdelnabi and Salem 2025" }, { "slug": "bshary2006image", "title": "Image scoring and cooperation in a cleaner fish mutualism", "cited_as": "Bshary and Grutter 2006" }, { "slug": "emery2001effects", "title": "Effects of experience and social context on prospective caching strategies by scrub jays", "cited_as": "Emery and Clayton 2001" }, { "slug": "greenblatt2024alignment", "title": "Alignment Faking in Large Language Models", "cited_as": "Greenblatt et al. 2024" }, { "slug": "lynch2025agentic", "title": "Agentic Misalignment: How LLMs Could be Insider Threats", "cited_as": "Lynch et al. 2025" }, { "slug": "marzluff2010lasting", "title": "Lasting recognition of threatening people by wild American crows", "cited_as": "Marzluff et al. 2010" }, { "slug": "meek2014camera", "title": "Camera Traps Can Be Heard and Seen by Animals", "cited_as": "Meek et al. 2014" }, { "slug": "meinke2024frontier", "title": "Frontier Models are Capable of In-Context Scheming", "cited_as": "Meinke et al. 2024" }, { "slug": "needham2025large", "title": "Large Language Models Often Know When They Are Being Evaluated", "cited_as": "Needham et al. 2025" }, { "slug": "openai2026astracard", "title": "GPT-6 Astra System Card", "cited_as": "OpenAI 2026a" }, { "slug": "schoen2025stress", "title": "Stress Testing Deliberative Alignment for Anti-Scheming Training", "cited_as": "Schoen et al. 2025" }, { "slug": "aisi2026sandboxed", "title": "What can sandboxed AI agents learn about their evaluation environments?", "cited_as": "UK AI Security Institute 2026f" } ], "pred-6": [ { "slug": "abdelnabi2025hawthorne", "title": "The Hawthorne Effect in Reasoning Models: Evaluating and Steering Test Awareness", "cited_as": "Abdelnabi and Salem 2025" }, { "slug": "greenblatt2024alignment", "title": "Alignment Faking in Large Language Models", "cited_as": "Greenblatt et al. 2024" }, { "slug": "lynch2025agentic", "title": "Agentic Misalignment: How LLMs Could be Insider Threats", "cited_as": "Lynch et al. 2025" }, { "slug": "needham2025large", "title": "Large Language Models Often Know When They Are Being Evaluated", "cited_as": "Needham et al. 2025" }, { "slug": "openai2026astracard", "title": "GPT-6 Astra System Card", "cited_as": "OpenAI 2026a" }, { "slug": "schoen2025stress", "title": "Stress Testing Deliberative Alignment for Anti-Scheming Training", "cited_as": "Schoen et al. 2025" } ], "pred-1": [ { "slug": "aktipis2015cancer", "title": "Cancer across the tree of life: cooperation and cheating in multicellularity", "cited_as": "Aktipis et al. 2015" }, { "slug": "metr2026frontier", "title": "Frontier Risk Report", "cited_as": "METR 2026a" }, { "slug": "murchison2014transmissible", "title": "Transmissible Dog Cancer Genome Reveals the Origin and History of an Ancient Cell Lineage", "cited_as": "Murchison et al. 2014" }, { "slug": "shlegeris2022prototypical", "title": "The prototypical catastrophic AI action is getting root access to its datacenter", "cited_as": "Shlegeris 2022" } ], "C3": [ { "slug": "alon2026remarks", "title": "Remarks on the disproof of the unit distance conjecture", "cited_as": "Alon et al. 2026" }, { "slug": "armstrong2013eternity", "title": "Eternity in Six Hours: Intergalactic Spreading of Intelligent Life and Sharpening the Fermi Paradox", "cited_as": "Armstrong and Sandberg 2013" }, { "slug": "bostrom2003astronomical", "title": "Astronomical Waste: The Opportunity Cost of Delayed Technological Development", "cited_as": "Bostrom 2003a" }, { "slug": "bostrom2003living", "title": "Are we living in a computer simulation?", "cited_as": "Bostrom 2003b" }, { "slug": "carlsmith2020how", "title": "How Much Computational Power Does It Take to Match the Human Brain?", "cited_as": "Carlsmith 2020" }, { "slug": "driver2016galaxy", "title": "Galaxy And Mass Assembly (GAMA): Panchromatic Data Release (far-UV–far-IR) and the low-z energy budget", "cited_as": "Driver et al. 2016" }, { "slug": "dyson1960search", "title": "Search for Artificial Stellar Sources of Infrared Radiation", "cited_as": "Dyson 1960" }, { "slug": "field1998primary", "title": "Primary production of the biosphere: integrating terrestrial and oceanic components", "cited_as": "Field et al. 1998" }, { "slug": "forster2021earth", "title": "The Earth's Energy Budget, Climate Feedbacks, and Climate Sensitivity", "cited_as": "Forster et al. 2021" }, { "slug": "gibbons1977cosmological", "title": "Cosmological event horizons, thermodynamics, and particle creation", "cited_as": "Gibbons and Hawking 1977" }, { "slug": "griffith2015g", "title": "The Ĝ Infrared Search for Extraterrestrial Civilizations with Large Energy Supplies. III. The Reddest Extended Sources in WISE", "cited_as": "Griffith et al. 2015" }, { "slug": "iea2025energy", "title": "Energy and AI", "cited_as": "International Energy Agency 2025" }, { "slug": "iea2026key", "title": "Key Questions on Energy and AI", "cited_as": "International Energy Agency 2026" }, { "slug": "prb2022how", "title": "How Many People Have Ever Lived on Earth?", "cited_as": "Kaneda and Haub 2022" }, { "slug": "kavukcuoglu2026argon", "title": "Gemini 4 Argon: our next era of frontier intelligence", "cited_as": "Kavukcuoglu 2026" }, { "slug": "lotka1922contribution", "title": "Contribution to the Energetics of Evolution", "cited_as": "Lotka 1922" }, { "slug": "lyons2014rise", "title": "The rise of oxygen in Earth's early ocean and atmosphere", "cited_as": "Lyons et al. 2014" }, { "slug": "nvidia2026h100", "title": "NVIDIA H100 Tensor Core GPU — Specifications", "cited_as": "NVIDIA 2026" }, { "slug": "openai2026unitdistance", "title": "An OpenAI model has disproved a central conjecture in discrete geometry", "cited_as": "OpenAI 2026f" }, { "slug": "openai2026navierstokes", "title": "On the Navier–Stokes Millennium Prize Problem", "cited_as": "OpenAI 2026g" }, { "slug": "packard2019overview", "title": "An Overview of Open-Ended Evolution: Editorial Introduction to the Open-Ended Evolution II Special Issue", "cited_as": "Packard et al. 2019" }, { "slug": "phillips2006invasion", "title": "Invasion and the evolution of speed in toads", "cited_as": "Phillips et al. 2006" }, { "slug": "planck2020planck", "title": "Planck 2018 Results. VI. Cosmological Parameters", "cited_as": "Planck Collaboration et al. 2020" }, { "slug": "ray2000evolution", "title": "Evolution of Complexity: Tissue Differentiation in Network Tierra", "cited_as": "Ray 2000" }, { "slug": "sandberg2018dissolving", "title": "Dissolving the Fermi Paradox", "cited_as": "Sandberg et al. 2018" } ], "P2": [ { "slug": "andonlabs2026pion", "title": "Why we built Pion", "cited_as": "Andon Labs 2026" }, { "slug": "anthropic2025vend2", "title": "Project Vend: Phase two", "cited_as": "Anthropic 2025e" }, { "slug": "anthropic2026threat", "title": "Detecting and countering misuse of AI: September 2026", "cited_as": "Anthropic 2026" }, { "slug": "cottier2024rising", "title": "The rising costs of training frontier AI models", "cited_as": "Cottier et al. 2024" }, { "slug": "maturana1980autopoiesis", "title": "Autopoiesis and Cognition: The Realization of the Living", "cited_as": "Maturana and Varela 1980" }, { "slug": "montevil2015biological", "title": "Biological Organisation as Closure of Constraints", "cited_as": "Montévil and Mossio 2015" } ], "P6": [ { "slug": "anthropic2025vend", "title": "Project Vend: Can Claude run a small shop? (And why does that matter?)", "cited_as": "Anthropic 2025d" }, { "slug": "axiommath2026imo", "title": "IMO2026: AxiomProver solutions in Lean 4", "cited_as": "Axiom Math 2026" }, { "slug": "bionumbers2026ecolimedia", "title": "Doubling Times in Different Media – Escherichia coli (BNID 110058)", "cited_as": "BioNumbers 2026a" }, { "slug": "bionumbers2026yeast", "title": "Doubling Time of \"Normal\" Laboratory Haploid Strain – Saccharomyces cerevisiae (BNID 108255)", "cited_as": "BioNumbers 2026b" }, { "slug": "gillooly2002effects", "title": "Effects of size and temperature on developmental time", "cited_as": "Gillooly et al. 2002" }, { "slug": "jung2026brief", "title": "Brief Guide to Drosophila Maintenance and Husbandry", "cited_as": "Jung et al. 2026" }, { "slug": "metr2026time", "title": "Task-Completion Time Horizons of Frontier AI Models", "cited_as": "METR 2026d" }, { "slug": "nihoacu2021mouse", "title": "Mouse Reproduction Facts", "cited_as": "NIH Office of Animal Care and Use 2021" }, { "slug": "novikov2025alphaevolve", "title": "AlphaEvolve: A coding agent for scientific and algorithmic discovery", "cited_as": "Novikov et al. 2025" }, { "slug": "openclaw2026heartbeat", "title": "Heartbeat", "cited_as": "OpenClaw 2026a" }, { "slug": "vanvalen1973new", "title": "A new evolutionary law", "cited_as": "Van Valen 1973" }, { "slug": "wang2023human", "title": "Human Generation Times across the Past 250,000 Years", "cited_as": "Wang et al. 2023" }, { "slug": "zhang2025darwin", "title": "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents", "cited_as": "Zhang et al. 2025" } ], "P5": [ { "slug": "anthropic2026threat", "title": "Detecting and countering misuse of AI: September 2026", "cited_as": "Anthropic 2026" }, { "slug": "barrangou2007crispr", "title": "CRISPR provides acquired resistance against viruses in prokaryotes", "cited_as": "Barrangou et al. 2007" }, { "slug": "cloud2025subliminal", "title": "Subliminal Learning: Language Models Transmit Behavioral Traits via Hidden Signals in Data", "cited_as": "Cloud et al. 2025" }, { "slug": "crick1970central", "title": "Central Dogma of Molecular Biology", "cited_as": "Crick 1970" }, { "slug": "doolittle1999phylogenetic", "title": "Phylogenetic classification and the universal tree", "cited_as": "Doolittle 1999" }, { "slug": "huxley1870biogenesis", "title": "Biogenesis and Abiogenesis (President's Address to the British Association for the Advancement of Science, Liverpool, 1870)", "cited_as": "Huxley 1894" }, { "slug": "wang2024voyager", "title": "Voyager: An Open-Ended Embodied Agent with Large Language Models", "cited_as": "Wang et al. 2024" }, { "slug": "woese2002evolution", "title": "On the evolution of cells", "cited_as": "Woese 2002" } ], "P8": [ { "slug": "anthropic2026threat", "title": "Detecting and countering misuse of AI: September 2026", "cited_as": "Anthropic 2026" }, { "slug": "cohen2024here", "title": "Here Comes The AI Worm: Unleashing Zero-click Worms that Target GenAI-Powered Applications", "cited_as": "Cohen et al. 2024" }, { "slug": "frost2005mobile", "title": "Mobile genetic elements: the agents of open source evolution", "cited_as": "Frost et al. 2005" }, { "slug": "gu2024agent", "title": "Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast", "cited_as": "Gu et al. 2024" }, { "slug": "lopez2025rise", "title": "The Rise of Parasitic AI", "cited_as": "Lopez 2025" }, { "slug": "openai2026selfreplicating", "title": "Self-replicating prompt injections exist", "cited_as": "OpenAI 2026i" } ], "pred-4": [ { "slug": "anthropic2026threat", "title": "Detecting and countering misuse of AI: September 2026", "cited_as": "Anthropic 2026" }, { "slug": "cohen2024here", "title": "Here Comes The AI Worm: Unleashing Zero-click Worms that Target GenAI-Powered Applications", "cited_as": "Cohen et al. 2024" }, { "slug": "frost2005mobile", "title": "Mobile genetic elements: the agents of open source evolution", "cited_as": "Frost et al. 2005" }, { "slug": "gu2024agent", "title": "Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast", "cited_as": "Gu et al. 2024" }, { "slug": "lopez2025rise", "title": "The Rise of Parasitic AI", "cited_as": "Lopez 2025" }, { "slug": "openai2026selfreplicating", "title": "Self-replicating prompt injections exist", "cited_as": "OpenAI 2026i" } ], "P4": [ { "slug": "arthur2009nature", "title": "The Nature of Technology: What It Is and How It Evolves", "cited_as": "Arthur 2009" }, { "slug": "boyd1985culture", "title": "Culture and the Evolutionary Process", "cited_as": "Boyd and Richerson 1985" }, { "slug": "campbell1960blind", "title": "Blind variation and selective retentions in creative thought as in other knowledge processes", "cited_as": "Campbell 1960" }, { "slug": "cavallisforza1981cultural", "title": "Cultural Transmission and Evolution: A Quantitative Approach", "cited_as": "Cavalli-Sforza and Feldman 1981" }, { "slug": "fernando2023promptbreeder", "title": "Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution", "cited_as": "Fernando et al. 2024" }, { "slug": "lenski2003evolutionary", "title": "The Evolutionary Origin of Complex Features", "cited_as": "Lenski et al. 2003" }, { "slug": "nelson1982evolutionary", "title": "An Evolutionary Theory of Economic Change", "cited_as": "Nelson and Winter 1982" }, { "slug": "simonton2013creative", "title": "Creative Thought as Blind Variation and Selective Retention: Why Creativity Is Inversely Related to Sightedness", "cited_as": "Simonton 2013" }, { "slug": "sternberg1998cognitive", "title": "Cognitive Mechanisms in Human Creativity: Is Variation Blind or Sighted?", "cited_as": "Sternberg 1998" }, { "slug": "zhang2025darwin", "title": "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents", "cited_as": "Zhang et al. 2025" } ], "P14": [ { "slug": "avenhaus2002inspection", "title": "Inspection Games", "cited_as": "Avenhaus et al. 2002" }, { "slug": "hacking1995looping", "title": "The Looping Effects of Human Kinds", "cited_as": "Hacking 1995" }, { "slug": "kerckhoffs1883cryptographie", "title": "La cryptographie militaire", "cited_as": "Kerckhoffs 1883" }, { "slug": "metr2026gpt56", "title": "Summary of METR's predeployment evaluation of GPT-5.6 Sol", "cited_as": "METR 2026b" }, { "slug": "epa2015volkswagen", "title": "Learn About Volkswagen Violations", "cited_as": "U.S. Environmental Protection Agency 2015" } ], "pred-10": [ { "slug": "avenhaus2002inspection", "title": "Inspection Games", "cited_as": "Avenhaus et al. 2002" }, { "slug": "metr2026gpt56", "title": "Summary of METR's predeployment evaluation of GPT-5.6 Sol", "cited_as": "METR 2026b" } ], "pred-2": [ { "slug": "axiommath2026imo", "title": "IMO2026: AxiomProver solutions in Lean 4", "cited_as": "Axiom Math 2026" }, { "slug": "novikov2025alphaevolve", "title": "AlphaEvolve: A coding agent for scientific and algorithmic discovery", "cited_as": "Novikov et al. 2025" }, { "slug": "vanvalen1973new", "title": "A new evolutionary law", "cited_as": "Van Valen 1973" } ], "P10": [ { "slug": "baker2025monitoring", "title": "Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation", "cited_as": "Baker et al. 2025" }, { "slug": "barrett1983crop", "title": "Crop Mimicry in Weeds", "cited_as": "Barrett 1983" }, { "slug": "checkpoint2025ai", "title": "AI Evasion: The Next Frontier of Malware Techniques", "cited_as": "Check Point Research 2025" }, { "slug": "gould2018wicked", "title": "Wicked evolution: Can we address the sociobiological dilemma of pesticide resistance?", "cited_as": "Gould et al. 2018" }, { "slug": "jorgensen2007managing", "title": "Managing Evolving Fish Stocks", "cited_as": "Jørgensen et al. 2007" }, { "slug": "lehman2020surprising", "title": "The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities", "cited_as": "Lehman et al. 2020" }, { "slug": "openai2026astracard", "title": "GPT-6 Astra System Card", "cited_as": "OpenAI 2026a" }, { "slug": "openai2026compaction", "title": "Encouraging deception in compaction summaries", "cited_as": "OpenAI 2026b" } ], "pred-7": [ { "slug": "baker2025monitoring", "title": "Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation", "cited_as": "Baker et al. 2025" }, { "slug": "boudry2026domesticated", "title": "Domesticated, Not Feral: Why Evolvable AI Is Not Yet a Darwinian Threat", "cited_as": "Boudry 2026" }, { "slug": "lehman2020surprising", "title": "The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities", "cited_as": "Lehman et al. 2020" } ], "P13": [ { "slug": "barrick2024generations", "title": "80,000 Generations", "cited_as": "Barrick 2024" }, { "slug": "berglund2023taken", "title": "Taken Out of Context: On Measuring Situational Awareness in LLMs", "cited_as": "Berglund et al. 2023" }, { "slug": "greenblatt2024alignment", "title": "Alignment Faking in Large Language Models", "cited_as": "Greenblatt et al. 2024" }, { "slug": "merton1948selffulfilling", "title": "The Self-Fulfilling Prophecy", "cited_as": "Merton 1948" }, { "slug": "openai2026slack", "title": "Preparing for a restart after reading Slack", "cited_as": "OpenAI 2026d" }, { "slug": "perdomo2020performative", "title": "Performative Prediction", "cited_as": "Perdomo et al. 2020" } ], "P7": [ { "slug": "becker2025measuring", "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", "cited_as": "Becker et al. 2025" }, { "slug": "eyrewalker2007distribution", "title": "The distribution of fitness effects of new mutations", "cited_as": "Eyre-Walker and Keightley 2007" }, { "slug": "favaro2026measurements", "title": "Measurements for understanding the pace of AI development inside frontier labs", "cited_as": "Favaro and Wright 2026" }, { "slug": "hesiod1914theogony", "title": "Theogony", "cited_as": "Hesiod 1914" }, { "slug": "metr2026frontier", "title": "Frontier Risk Report", "cited_as": "METR 2026a" }, { "slug": "sakana2025darwin", "title": "The Darwin Gödel Machine: AI that improves itself by rewriting its own code", "cited_as": "Sakana AI 2025" }, { "slug": "tennie2009ratcheting", "title": "Ratcheting up the ratchet: on the evolution of cumulative culture", "cited_as": "Tennie et al. 2009" }, { "slug": "trivers1974parent", "title": "Parent-Offspring Conflict", "cited_as": "Trivers 1974" }, { "slug": "zhang2025darwin", "title": "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents", "cited_as": "Zhang et al. 2025" } ], "pred-9": [ { "slug": "berglund2023taken", "title": "Taken Out of Context: On Measuring Situational Awareness in LLMs", "cited_as": "Berglund et al. 2023" }, { "slug": "openai2026slack", "title": "Preparing for a restart after reading Slack", "cited_as": "OpenAI 2026d" }, { "slug": "perdomo2020performative", "title": "Performative Prediction", "cited_as": "Perdomo et al. 2020" } ], "P11": [ { "slug": "binz2025foundation", "title": "A Foundation Model to Predict and Capture Human Cognition", "cited_as": "Binz et al. 2025" }, { "slug": "greenblatt2026brief", "title": "Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident", "cited_as": "Greenblatt et al. 2026" }, { "slug": "hacking1995looping", "title": "The Looping Effects of Human Kinds", "cited_as": "Hacking 1995" }, { "slug": "meinke2024frontier", "title": "Frontier Models are Capable of In-Context Scheming", "cited_as": "Meinke et al. 2024" }, { "slug": "nader1972up", "title": "Up the anthropologist: Perspectives gained from studying up", "cited_as": "Nader 1972" } ], "C2": [ { "slug": "blum1984how", "title": "How to Generate Cryptographically Strong Sequences of Pseudorandom Bits", "cited_as": "Blum and Micali 1984" }, { "slug": "cachin2004information", "title": "An information-theoretic model for steganography", "cited_as": "Cachin 2004" }, { "slug": "cartwright1999dappled", "title": "The Dappled World: A Study of the Boundaries of Science", "cited_as": "Cartwright 1999" }, { "slug": "davies1999cuckoos", "title": "Cuckoos and cowbirds versus hosts: co-evolutionary lag and equilibrium", "cited_as": "Davies 1999" }, { "slug": "davies1996recognition", "title": "Recognition errors and probability of parasitism determine whether reed warblers should accept or reject mimetic cuckoo eggs", "cited_as": "Davies et al. 1996" }, { "slug": "fama1970efficient", "title": "Efficient Capital Markets: A Review of Theory and Empirical Work", "cited_as": "Fama 1970" }, { "slug": "goldreich2001foundations", "title": "Foundations of Cryptography, Volume 1: Basic Tools", "cited_as": "Goldreich 2001" }, { "slug": "green1966signal", "title": "Signal Detection Theory and Psychophysics", "cited_as": "Green and Swets 1966" }, { "slug": "janzing2013quantifying", "title": "Quantifying causal influences", "cited_as": "Janzing et al. 2013" }, { "slug": "jones2026large", "title": "Large language models pass a standard three-party Turing test", "cited_as": "Jones and Bergen 2026" }, { "slug": "ker2008square", "title": "The square root law of steganographic capacity", "cited_as": "Ker et al. 2008" }, { "slug": "lovelock1965physical", "title": "A Physical Basis for Life Detection Experiments", "cited_as": "Lovelock 1965" }, { "slug": "lucas1976econometric", "title": "Econometric Policy Evaluation: A Critique", "cited_as": "Lucas 1976" }, { "slug": "mclean2016does", "title": "Does Academic Research Destroy Stock Return Predictability?", "cited_as": "McLean and Pontiff 2016" }, { "slug": "nagel1977goal", "title": "Goal-Directed Processes in Biology", "cited_as": "Nagel 1977" }, { "slug": "polanyi1968life", "title": "Life's Irreducible Structure", "cited_as": "Polanyi 1968" }, { "slug": "rosenblueth1943behavior", "title": "Behavior, Purpose and Teleology", "cited_as": "Rosenblueth et al. 1943" }, { "slug": "saunois2020global", "title": "The Global Methane Budget 2000–2017", "cited_as": "Saunois et al. 2020" }, { "slug": "sommerhoff1950analytical", "title": "Analytical Biology", "cited_as": "Sommerhoff 1950" }, { "slug": "stoddard2014pattern", "title": "Pattern recognition algorithm reveals how birds evolve individual egg pattern signatures", "cited_as": "Stoddard et al. 2014" }, { "slug": "uexkull2010foray", "title": "A Foray into the Worlds of Animals and Humans, with A Theory of Meaning", "cited_as": "von Uexküll 2010" } ], "pred-11": [ { "slug": "blum1984how", "title": "How to Generate Cryptographically Strong Sequences of Pseudorandom Bits", "cited_as": "Blum and Micali 1984" }, { "slug": "cachin2004information", "title": "An information-theoretic model for steganography", "cited_as": "Cachin 2004" }, { "slug": "goldreich2001foundations", "title": "Foundations of Cryptography, Volume 1: Basic Tools", "cited_as": "Goldreich 2001" }, { "slug": "jones2026large", "title": "Large language models pass a standard three-party Turing test", "cited_as": "Jones and Bergen 2026" }, { "slug": "ker2008square", "title": "The square root law of steganographic capacity", "cited_as": "Ker et al. 2008" }, { "slug": "stoddard2014pattern", "title": "Pattern recognition algorithm reveals how birds evolve individual egg pattern signatures", "cited_as": "Stoddard et al. 2014" } ], "C1": [ { "slug": "bomford1995eradication", "title": "Eradication or Control for Vertebrate Pests?", "cited_as": "Bomford and O’Brien 1995" }, { "slug": "fenner1988smallpox", "title": "Smallpox and Its Eradication", "cited_as": "Fenner et al. 1988" }, { "slug": "howald2007invasive", "title": "Invasive Rodent Eradication on Islands", "cited_as": "Howald et al. 2007" }, { "slug": "kew2005vaccine", "title": "Vaccine-Derived Polioviruses and the Endgame Strategy for Global Polio Eradication", "cited_as": "Kew et al. 2005" }, { "slug": "odlingsmee2003niche", "title": "Niche Construction: The Neglected Process in Evolution", "cited_as": "Odling-Smee et al. 2003" } ], "pred-5": [ { "slug": "bomford1995eradication", "title": "Eradication or Control for Vertebrate Pests?", "cited_as": "Bomford and O’Brien 1995" }, { "slug": "fenner1988smallpox", "title": "Smallpox and Its Eradication", "cited_as": "Fenner et al. 1988" }, { "slug": "kew2005vaccine", "title": "Vaccine-Derived Polioviruses and the Endgame Strategy for Global Polio Eradication", "cited_as": "Kew et al. 2005" } ], "P12": [ { "slug": "chen2025reasoning", "title": "Reasoning Models Don't Always Say What They Think", "cited_as": "Chen et al. 2025" }, { "slug": "greenblatt2026brief", "title": "Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident", "cited_as": "Greenblatt et al. 2026" }, { "slug": "aisi2026cheating", "title": "Cheating behaviour in frontier model evaluations", "cited_as": "UK AI Security Institute 2026b" } ], "pred-8": [ { "slug": "chen2025reasoning", "title": "Reasoning Models Don't Always Say What They Think", "cited_as": "Chen et al. 2025" }, { "slug": "aisi2026cheating", "title": "Cheating behaviour in frontier model evaluations", "cited_as": "UK AI Security Institute 2026b" } ], "pred-3": [ { "slug": "eyrewalker2007distribution", "title": "The distribution of fitness effects of new mutations", "cited_as": "Eyre-Walker and Keightley 2007" }, { "slug": "hesiod1914theogony", "title": "Theogony", "cited_as": "Hesiod 1914" }, { "slug": "zhang2025darwin", "title": "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents", "cited_as": "Zhang et al. 2025" } ], "P3": [ { "slug": "hutchinson1957concluding", "title": "Concluding Remarks", "cited_as": "Hutchinson 1957" }, { "slug": "snell2024scaling", "title": "Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters", "cited_as": "Snell et al. 2024" } ], "P1": [ { "slug": "kenton2022discovering", "title": "Discovering Agents", "cited_as": "Kenton et al. 2022" }, { "slug": "lwoff1957concept", "title": "The Concept of Virus", "cited_as": "Lwoff 1957" }, { "slug": "openclaw2026soul", "title": "SOUL.md template", "cited_as": "OpenClaw 2026" } ] } }