{"ontology":"https://layer.re/machine/schema.json","epistemic_model":"evidence-backed-temporal-layer-graph","canonical":true,"type":"civilization_snapshot","snapshot":{"id":"civ-2022q1-reconstructed","as_of":"2022-03-31 23:59:59.000","summary":"By the end of Q1 2022, the emerging AI/civilization stack was centered on research-driven model progress, dedicated accelerator compute, high-bandwidth networking, software interfaces and growing semiconductor strategic policy. InstructGPT operationalized human-feedback post-training; chain-of-thought and AlphaCode showed that prompting and code generation could expand model capability surfaces; Meta RSC and NVIDIA H100 made specialized infrastructure increasingly explicit; the EU Chips Act proposal, U.S. export controls and NIST AI RMF draft showed governance and geopolitics coupling to chips and AI. In this evidence set, persistent Agents and agent-specific Permissions were still precursor/underdeveloped rather than independent operational layers.","cross_layer_changes":["Dedicated AI compute became visibly coupled to chips, networks, storage and model research.","Human feedback became a production input into deployed model behavior.","Prompting and code generation expanded the software-facing capability surface of large models.","Open-model access became a more material ecosystem path.","Semiconductor manufacturing and access became explicit governance and geopolitical concerns.","Formal AI risk-management processes advanced before the later generative-AI boom."],"dependencies":["research -> models","human feedback -> models","chips + networks -> compute","compute -> model research and training","models -> software capability","manufacturing + governance + geopolitics -> chip availability"],"constraints":["accelerator access and cost","network and system scale","training data/model allocation","model reliability and controllability","semiconductor supply concentration","policy implementation lag"],"emerging_capabilities":["instruction-following deployed language models","prompt-elicited multi-step reasoning","competition-level code generation","publicly accessible open large language models","transformer-specific accelerator systems"],"possible_bottlenecks":["specialized compute access","semiconductor manufacturing and supply","high-bandwidth interconnect","training efficiency and data allocation","model supervision and reliability"],"uncertainties":["This reconstruction is intentionally selective and stack-focused rather than a complete archive of Q1 2022 AI.","The absence of a mapped layer event does not prove that the layer was irrelevant in all domains.","Persistent agents and agent-specific permissions are treated as not yet independently evidenced in this source set, not as nonexistent.","Energy is not mapped as a distinct AI bottleneck in Q1 because the selected evidence does not support that claim strongly enough.","Later importance of Q1 research methods is retrospective and should not be attributed as contemporaneous consensus."],"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.399Z","updated_at":"2026-10-01T18:05:36.399Z"},"claims":[{"id":"claim-q1-compute-coupling","statement":"By the end of Q1 2022, frontier AI model development was already materially coupled to specialized accelerator compute and high-bandwidth networking.","epistemic_status":"supported_inference","confidence":0.95,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-compute-2022q1","competing_interpretation":"Large model progress could still be framed primarily as an algorithmic/research story because only a small number of organizations operated infrastructure at this scale.","uncertainty":"The degree to which dedicated supercomputing was necessary varied by model family and organization.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.400Z","updated_at":"2026-10-01T18:05:36.400Z"},{"id":"claim-q1-governance-geopolitics","statement":"Q1 2022 shows governance and geopolitics coupling to the technology stack through AI risk frameworks, semiconductor industrial policy and export controls.","epistemic_status":"supported_inference","confidence":0.95,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-governance-2022q1","competing_interpretation":"These policy streams had different motivations and should not be collapsed into a single AI-governance agenda.","uncertainty":"The EU Chips Act was still a proposal, the NIST AI RMF was a draft, and Russia export controls were not AI-specific.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.401Z","updated_at":"2026-10-01T18:05:36.401Z"},{"id":"claim-q1-models-software","statement":"Q1 2022 established a credible early path from large language models into software work through prompting and code-generation systems, without yet demonstrating reliable autonomous software agents.","epistemic_status":"supported_inference","confidence":0.92,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-software-2022q1","competing_interpretation":"AlphaCode's constrained evaluation and heavy sampling/filtering may make it a research demonstration rather than a direct precursor to production software engineering.","uncertainty":"Transfer from competition problems to real software development was unproven.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.400Z","updated_at":"2026-10-01T18:05:36.400Z"},{"id":"claim-q1-open-access","statement":"Open and publicly accessible large-model ecosystems were becoming materially relevant in Q1 2022, although they had not reached the scale or capability of the largest closed frontier systems.","epistemic_status":"supported_inference","confidence":0.88,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-open-source-2022q1","competing_interpretation":"A 20B model may represent community access without constituting a strategically independent frontier-model layer.","uncertainty":"Definitions of open source for models and the exact timing of downloadable weight availability require care.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.401Z","updated_at":"2026-10-01T18:05:36.401Z"},{"id":"claim-q1-posttraining","statement":"Q1 2022 contains strong evidence that human-feedback post-training had become an operational production mechanism for shaping deployed language-model behavior.","epistemic_status":"supported_inference","confidence":0.98,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-models-2022q1","competing_interpretation":"The core RLHF technique predates Q1 2022, so the event is better interpreted as maturation and deployment rather than emergence from zero.","uncertainty":"The durability and generality of the approach across providers were not yet established.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.400Z","updated_at":"2026-10-01T18:05:36.400Z"},{"id":"claim-q1-preagent","statement":"In this Q1 2022 reconstruction, Agents and agent-specific Permissions are better represented as precursor/underdeveloped layers than as independent operational layers.","epistemic_status":"supported_inference","confidence":0.82,"civilization_snapshot_id":"civ-2022q1-reconstructed","layer_state_snapshot_id":"state-agents-2022q1","competing_interpretation":"Earlier autonomous-agent and reinforcement-learning research could justify a broader historical definition of the Agents layer.","uncertainty":"This judgment is scoped to the present Layer.re ontology and the curated material events selected for the AI/civilization stack.","mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.402Z","updated_at":"2026-10-01T18:05:36.402Z"}],"layer_states":[{"id":"state-ai-2022q1","layer_id":"layer-ai","as_of":"2022-03-31 23:59:59.000","summary":"AI progress in Q1 2022 was centered on large models, instruction-following post-training, prompting methods, code generation and increasingly specialized infrastructure; persistent agents were not yet evidenced as a separate operational layer.","dependencies":["layer-models","layer-research","layer-compute","layer-software"],"constraints":["model reliability","training cost","compute access","limited controllability"],"emerging_capabilities":["instruction-following deployed models","prompt-elicited multi-step reasoning","competitive code generation"],"underdeveloped_components":["persistent autonomous action","mature external-tool orchestration"],"uncertainties":["selected evidence is LLM-heavy and does not cover all AI subfields"],"confidence":0.91,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.387Z","updated_at":"2026-10-01T18:05:36.387Z","layer_slug":"ai","layer_name":"AI"},{"id":"state-agents-2022q1","layer_id":"layer-agents","as_of":"2022-03-31 23:59:59.000","summary":"Within the selected Q1 2022 evidence, models could follow instructions, reason under prompting and generate code, but persistent tool-using autonomous agents are not yet supported as an independent operational layer.","dependencies":["layer-models","layer-software"],"constraints":["reliable long-horizon execution","tool use","persistent state","delegation controls"],"emerging_capabilities":["code/problem-solving systems as a precursor, not yet general agents"],"underdeveloped_components":["persistent autonomy","general tool orchestration","background execution"],"uncertainties":["absence from this curated source set is not proof that no agent research existed anywhere"],"confidence":0.78,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.398Z","updated_at":"2026-10-01T18:05:36.398Z","layer_slug":"agents","layer_name":"Agents"},{"id":"state-chips-2022q1","layer_id":"layer-chips","as_of":"2022-03-31 23:59:59.000","summary":"Semiconductors were simultaneously becoming more AI-specialized and more strategically governed: H100 targeted transformer workloads while the EU and U.S. treated chip access and supply as strategic policy concerns.","dependencies":["layer-manufacturing","layer-software","layer-networks","layer-governance","layer-geopolitics"],"constraints":["supply concentration","manufacturing capacity","export controls","availability lag from announcement to deployment"],"emerging_capabilities":["transformer-specific acceleration","higher-bandwidth scaled GPU systems"],"underdeveloped_components":["regional manufacturing resilience"],"uncertainties":["policy proposals and product announcements did not yet guarantee later capacity or adoption"],"confidence":0.96,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.394Z","updated_at":"2026-10-01T18:05:36.394Z","layer_slug":"chips","layer_name":"Chips"},{"id":"state-compute-2022q1","layer_id":"layer-compute","as_of":"2022-03-31 23:59:59.000","summary":"Dedicated AI compute was already becoming a strategic operational layer, exemplified by Meta RSC and by NVIDIA's next-generation H100/DGX platform announcement.","dependencies":["layer-chips","layer-networks","layer-software","storage infrastructure"],"constraints":["accelerator availability","interconnect bandwidth","capital intensity","training duration"],"emerging_capabilities":["multi-thousand-GPU research clusters","next-generation DGX/SuperPOD systems"],"underdeveloped_components":["broad access outside well-capitalized organizations"],"uncertainties":["announcement performance and future buildout claims require later verification"],"confidence":0.96,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.394Z","updated_at":"2026-10-01T18:05:36.394Z","layer_slug":"compute","layer_name":"Compute"},{"id":"state-geopolitics-2022q1","layer_id":"layer-geopolitics","as_of":"2022-03-31 23:59:59.000","summary":"Technology supply chains were already explicit geopolitical objects: Europe framed semiconductor dependence as a sovereignty/resilience problem and U.S.-led export controls restricted Russia's technology access.","dependencies":["layer-chips","layer-manufacturing","layer-governance","layer-software"],"constraints":["cross-border dependencies","export-control exposure","concentrated semiconductor supply"],"emerging_capabilities":["coordinated technology restrictions","regional semiconductor industrial strategy"],"underdeveloped_components":["stable cross-border technology regimes"],"uncertainties":["Russia controls were not AI-specific; later China-focused AI-chip controls must not be projected backward"],"confidence":0.94,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.397Z","updated_at":"2026-10-01T18:05:36.397Z","layer_slug":"geopolitics","layer_name":"Geopolitics"},{"id":"state-governance-2022q1","layer_id":"layer-governance","as_of":"2022-03-31 23:59:59.000","summary":"Governance was becoming independently operational around both AI risk management and strategic semiconductor capacity, using frameworks, industrial policy and export controls.","dependencies":["layer-ai","layer-chips","layer-manufacturing","layer-geopolitics","institutional capacity"],"constraints":["nonbinding AI risk processes","policy implementation lag","jurisdictional fragmentation"],"emerging_capabilities":["structured AI risk-management drafts","semiconductor resilience policy","technology export controls"],"underdeveloped_components":["mature generative-AI-specific rules","agent liability regimes"],"uncertainties":["NIST AI RMF was a draft and the EU Chips Act was a proposal during Q1"],"confidence":0.95,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.396Z","updated_at":"2026-10-01T18:05:36.396Z","layer_slug":"governance","layer_name":"Governance"},{"id":"state-human-2022q1","layer_id":"layer-human","as_of":"2022-03-31 23:59:59.000","summary":"Humans were becoming more directly embedded in model behavior through preference labeling and feedback, while institutions were beginning to formalize AI risk to individuals and society.","dependencies":["layer-models","layer-governance","human labelers and users"],"constraints":["scalable supervision","bias and representativeness","uncertain social impacts"],"emerging_capabilities":["human-feedback post-training","structured societal-risk framing"],"underdeveloped_components":["scalable oversight for more capable systems"],"uncertainties":["human preference data may not represent broader populations or future use contexts"],"confidence":0.89,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.397Z","updated_at":"2026-10-01T18:05:36.397Z","layer_slug":"human","layer_name":"Human"},{"id":"state-manufacturing-2022q1","layer_id":"layer-manufacturing","as_of":"2022-03-31 23:59:59.000","summary":"Semiconductor manufacturing capacity was treated as a strategic prerequisite for digital and future AI capability, especially in the proposed European Chips Act.","dependencies":["layer-research","capital investment","supply chains","layer-governance"],"constraints":["fab lead times","geographic concentration","capital intensity","equipment dependencies"],"emerging_capabilities":["public-private capacity-building programs proposed"],"underdeveloped_components":["rapid regional capacity expansion"],"uncertainties":["Q1 policy intent should not be confused with later realized fabrication capacity"],"confidence":0.89,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.397Z","updated_at":"2026-10-01T18:05:36.397Z","layer_slug":"manufacturing","layer_name":"Manufacturing"},{"id":"state-models-2022q1","layer_id":"layer-models","as_of":"2022-03-31 23:59:59.000","summary":"Large language models were moving beyond raw next-token pretraining through human-feedback post-training, prompting methods and compute-efficient scaling research.","dependencies":["layer-research","layer-compute","layer-chips","layer-software","layer-human"],"constraints":["training compute","data allocation","reliability and truthfulness","deployment cost"],"emerging_capabilities":["instruction following","reasoning elicitation by prompting","code generation"],"underdeveloped_components":["robust general reasoning","reliable long-horizon planning"],"uncertainties":["later importance of these methods is retrospective; contemporaneous adoption varied"],"confidence":0.95,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.388Z","updated_at":"2026-10-01T18:05:36.388Z","layer_slug":"models","layer_name":"Models"},{"id":"state-networks-2022q1","layer_id":"layer-networks","as_of":"2022-03-31 23:59:59.000","summary":"High-bandwidth interconnects were already a necessary part of scaling AI compute across thousands of accelerators, visible in RSC InfiniBand and H100 NVLink system design.","dependencies":["layer-chips","layer-compute"],"constraints":["bandwidth","latency","cluster scale","system reliability"],"emerging_capabilities":["large non-oversubscribed training fabrics","NVLink-scaled DGX systems"],"underdeveloped_components":["portable cross-vendor scale-up standards"],"uncertainties":["vendor architecture claims are stronger than independent comparative evidence"],"confidence":0.91,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.395Z","updated_at":"2026-10-01T18:05:36.395Z","layer_slug":"networks","layer_name":"Networks"},{"id":"state-open-source-2022q1","layer_id":"layer-open-source","as_of":"2022-03-31 23:59:59.000","summary":"Open-model infrastructure was maturing from smaller community models toward publicly accessible multi-billion-parameter systems such as GPT-NeoX-20B.","dependencies":["layer-software","layer-compute","layer-research"],"constraints":["GPU access","serving cost","gap to largest closed models"],"emerging_capabilities":["publicly accessible large-model APIs","open training/model software"],"underdeveloped_components":["frontier-scale parity","uniform definitions of model openness"],"uncertainties":["Feb 2 public access preceded the announced full checkpoint release"],"confidence":0.86,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.396Z","updated_at":"2026-10-01T18:05:36.396Z","layer_slug":"open-source","layer_name":"Open source"},{"id":"state-permissions-2022q1","layer_id":"layer-permissions","as_of":"2022-03-31 23:59:59.000","summary":"The selected Q1 evidence shows API access controls, privacy processes and institutional governance, but not a distinct agent-permission infrastructure for delegated autonomous action.","dependencies":["layer-software","layer-governance"],"constraints":["no standardized delegation semantics evidenced","no agent runtime enforcement layer evidenced"],"emerging_capabilities":[],"underdeveloped_components":["agent-specific authorization","portable delegated permissions","runtime policy enforcement"],"uncertainties":["general computing permissions existed, but their later agent-layer significance should not be projected backward"],"confidence":0.72,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.398Z","updated_at":"2026-10-01T18:05:36.398Z","layer_slug":"permissions","layer_name":"Permissions"},{"id":"state-research-2022q1","layer_id":"layer-research","as_of":"2022-03-31 23:59:59.000","summary":"Research was a primary upstream driver of model capability, with Q1 work on RLHF deployment, prompting, code generation and compute-optimal scaling quickly connecting to deployable systems.","dependencies":["layer-compute","layer-models","academic and industrial research organizations"],"constraints":["experimental compute requirements","benchmark representativeness","rapidly changing baselines"],"emerging_capabilities":["RLHF methods","chain-of-thought prompting","compute-optimal scaling analysis"],"underdeveloped_components":["standardized independent evaluation"],"uncertainties":["some preprints were revised later; only Q1-supported claims should be attributed to the quarter"],"confidence":0.94,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.388Z","updated_at":"2026-10-01T18:05:36.388Z","layer_slug":"research","layer_name":"Research"},{"id":"state-software-2022q1","layer_id":"layer-software","as_of":"2022-03-31 23:59:59.000","summary":"Software became more tightly coupled to model capability through prompting, code generation, APIs and accelerator software stacks.","dependencies":["layer-models","layer-open-source","layer-chips","layer-compute"],"constraints":["model reliability","deployment complexity","hardware specificity"],"emerging_capabilities":["LLM APIs as application substrate","model-generated code","prompt-based capability elicitation"],"underdeveloped_components":["general autonomous software engineering"],"uncertainties":["AlphaCode performance does not imply production software-agent reliability"],"confidence":0.91,"previous_snapshot_id":null,"mapping_mode":"retrospective","reconstructed_at":"2026-10-01T17:34:32.000Z","created_at":"2026-10-01T18:05:36.395Z","updated_at":"2026-10-01T18:05:36.395Z","layer_slug":"software","layer_name":"Software"}]}