A Further Exploration of the AGI Delusion

Main point: the author argues that scaling (more tokens, parameters, compute) alone will not produce AGI and that current LLM-centric approaches are functionally hollow. The piece advocates shifting toward architectures that model unconscious, experiential memory—‘memory implants’/experience packs, digital twins, and neuro-secure protocols—to enable pragmatic decision-making and human-machine symbiosis; these ideas imply new product and identity markets but are speculative and long-term.

Analysis

The article reframes the AI battleground from “bigger models + more data” to “engineered experience” — a shift that creates a new value chain: memory creation, provenance, and identity assurance. That pushes economic rents away from raw compute and generic LLM API sellers toward firms that can certify, store, and act on personally or institutionally curated experience-graphs; those assets are durable and monetizable because they are costly to replicate and legally sensitive.

Second-order effects: cloud/storage providers and enterprise integrators will face both revenue opportunities (subscription fees, vaulting, indexing) and liability risk (false memories, impersonation, regulatory subpoenas), which means insurance, compliance tooling, and audit logs become revenue drivers on a multi-year cadence. At the hardware layer, demand will bifurcate — commodity transformer compute growth slows if experience-based agents use different architectures, but specialized inference/neuromorphic and high-throughput memory systems could see concentrated capex from defense, healthcare, and identity-sensitive industries within 1–4 years.

Timing and reversal: memory-pack monetization and “neuro-secure” standards are months-to-years plays; expect measurable enterprise pilots in 6–18 months and standards/regulatory skirmishes over identity in 12–36 months. The thesis breaks if (a) transformer scaling achieves robust, low-cost real-world agency within 12 months, or (b) regulators outlaw synthetic biographies or assign strict legal personhood/liability to digital twins — either would reallocate rents back to scale-focused incumbents or collapse the experience-pack market overnight.

Key Decisions for Investors

• Long OKTA (OKTA) — buy shares or 12–18 month ITM calls on pullbacks. Rationale: identity control and sovereign digital-twin plumbing are natural buyers of Okta-like tech; expected 30–80% upside if enterprise pilots convert to contracts within 12–24 months. Risk: large cloud outage or single-vendor replacement could compress returns by ~30%.
• Long CrowdStrike (CRWD) — accumulate over 3–12 months or buy Jan-2027 calls sized to 3–5% of portfolio. Rationale: ‘neuro-secure’ audit/logging and breach-prevention services become mandatory for experience-pack deployments; potential 2x upside if ARR growth accelerates. Risk: macro-driven multiple compression or missed execution could halve gains.
• Paired trade: Long Palantir (PLTR) / Short C3.ai (AI) — 12–24 month pair (equal dollar exposure). Rationale: PLTR’s data integration and mission-oriented data fabrics win reality-recorder projects; C3.ai is a candidate to re-rate down if scale-only AI revenue disappoints. Target asymmetric payoff: 40–100% on long if adoption occurs while short trims 30–60% on valuation compression. Risk: broad AI re-rating up would hurt the short leg.
• Long NVIDIA (NVDA) selective calls 3–12 months (smaller size) — hedge for continued high-end inference demand. Rationale: even if architecture shifts, specialized accelerators and memory-heavy systems keep near-term NVDA revenue resilient; expect 20–50% upside in a continued AI capex cycle. Risk: a rapid pivot to low-cost custom silicon or secular compute downturn could erase short-term gains.