Prime Radiant/Machine Cards
LMDawnclass card

InfoSubstrate GenAI (LM-EXTRAP)

culture pace layer · 2022–ongoing

lifespan: 200 yrs

Terminal class card in the InfoSubstrate lineage: the LM-Dawn epistemic substrate that succeeds the DM attention-platform era (InfoSubstrate-SocialPlatform-1995) when generative AI displaces broadcast/social-platform attention-economy logic with hybrid human-machine information production. Distinguishing features vs. predecessors: (1) motor is absent — no progress-telos drives the substrate; information production is constituted by feedback loops between human intent, LLM synthesis, retrieval infrastructure, and multi-agent verification cycles; (2) symbol-substrate is reflexive — LLMs trained on human text produce outputs that re-enter the training corpus, creating self-modifying epistemic loops (Bender et al. 2021: "stochastic parrots" critique crystallizes the recursive risk); (3) information has hybrid-provenance — every output is potentially human-authored, LLM-synthesized, human-curated-LLM, or multi-agent-verified; provenance is structurally opaque at point of consumption; (4) epistemic-trust is post-broadcast and post-platform — the shared public-agenda (MM newspaper- broadcast OPP) and the algorithmic engagement-feed (DM platform OPP) are both sublimated; information environments become personalised, agent-mediated, and verification-dependent rather than authority-dependent. Typology break from DM SocialPlatform fires on three patterns: (1) substrate_replacement — the production infrastructure shifts from incorporeal DM-platform (server + social-graph + ad-auction) to incorporeal LM-hybrid (transformer inference stack + retrieval index + human-LLM workflow); (2) input_set_replacement — dominant inputs shift from UGC + behavioral data (DM) to natural-language prompts + retrieval corpora + human feedback signals; (3) coupling_typology_shift — the platform OPP (ad-auction monetises attention) is replaced by API-subscription, RAG-services, and agentic-workflow monetisation. emergence_subtype: hybrid-agentic (LLM-assisted co-production between humans and models) — does NOT map cleanly to crowdsourced or meritocratic_hierarchy; stored here in description per schema constraint (EmergenceSubtype enum only contains crowdsourced/meritocratic_hierarchy; hybrid-agentic falls outside closed enum). [STUB-substrate-enum-gap]: the substrate type "institutional" is MISSING from SubstrateType enum in v0.1; the epistemic-governance dimension (oversight bodies, audit boards, provenance standards) is encoded in description, not substrate list. Confidence ≤ 0.45 throughout: this is LM-EXTRAP forward-extrapolation only. All quantitative state-variable values are [EXTRAP]; framing draws on [CANON] sources (Bender et al. 2021; Marcus & Davis 2019; Goldfarb & Tucker 2019; Wave-0 LM era frame; Wave-6 InfoSubstrate lineage).

Machine type

incorporeal

Plasticity

plastic

Substrate

incorporeal cognitive semiotic

Wave source

phase-1-batch3k-audit-resolution-2026-05-26

Inputs

  • Natural-language prompts (human intent, queries, instructions)
  • Retrieval corpora (pre-internet + internet text; scientific literature; code)
  • Human feedback signals (RLHF, RLAIF, evals, red-teaming)
  • GPU compute electricity (consumed cloud inference substrate)

Outputs

  • Hybrid-provenance information artifacts (human-LLM co-authored)
  • Reflexive epistemic corpus (LLM outputs re-entering training data)
  • Agentic workflow substrate (multi-agent task execution layer)

Landscape pressures

  • epistemic_trust_collapse_pressure (85% intensity)
  • synthetic_content_provenance_crisis (80% intensity)
  • compute_concentration_regulatory_pressure (70% intensity)
  • reflexive_corpus_contamination (65% intensity)

Intra-era couplings

Cross-era couplings

State variables

coordination_yield_index
0.32
EXTRAP
capture_resistance_index
0.42
EXTRAP
self_organized_criticality_proximity
0.65
EXTRAP
liveness_temporal_coupling
0.45
ontological_doubt_prevalence
0.72
EXTRAP
real_virtuality_saturation
0.55
EXTRAP
proletarianization_risk
0.72
machine_lifespan
200
regime
chaotic
CANON
gravitational_weight
0.30
EXTRAP
black_box_depth
12
EXTRAP

Phase snapshots

LM-Dawn2022–2028chaotic
LM-Dawn2030–2050chaotic

Notable instances

  • Perplexity AI (2022) (2022) — [EXTRAP] Retrieval-augmented generation as primary UX (search-meets-LLM); canonical early instance of the RAG-first GenA…
  • Midjourney (2022) (2022) — [EXTRAP] Generative image substrate; extends GenAI beyond text into visual-semiotic production. Hybrid-provenance logic …
  • Retrieval-Augmented Generation (RAG) pattern (2020) — [EXTRAP] RAG as a class-pattern (not a specific product): attaches LLM inference to live retrieval indexes, partially re…
  • Fediverse-LLM hybrid instances (2024+) (2024) — [EXTRAP] Experimental instances coupling ActivityPub-based federated social networks (Mastodon, Bluesky) with embedded L…

Sources

  • Bender, Gebru, McMillan-Major, Mitchell (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? · 90%
  • Marcus, Gary; Davis, Ernest (2019). Rebooting AI: Building Artificial Intelligence We Can Trust · 85%
  • Goldfarb, Avi; Tucker, Catherine (2019). Digital Economics · 85%
  • Wave-0 (2026). world-machines-eras-frame/findings.md (LM era definitions)
  • Wave-6 Substitution Lineage (2026). machine-substitution-lineage/findings.md §Chain-5 (InfoSubstrate)
  • Castells, Manuel (1996). The Rise of the Network Society · 85%