brain/
← all contexts
threadlast active 2026-05-15

autonomous-driving

0 mechanisms4 entities3 concepts3 questions2 sources
Recent activity
ingest2026-05-15 15:00
2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts

Multi-context source (vault/sources/) — also tagged for threads/artificial-intelligence and projects/career. Speaker-aware ingest (named speakers Andrej Karpathy + Dwarkesh Patel via AssemblyAI Speaker Identification). Self-driving-relevant content is concentrated in the "Why self driving took so long" segment (Karpathy ran Tesla Autopilot 2017–2022).

  • Updated andrej-karpathy (cross-context — physical home: vault/threads/artificial-intelligence/wiki/entities/, tag thread/autonomous-driving appended, +7 quotes attributed). The artificial-intelligence ingest created the page; this ingest adds the self-driving-specific quotes.
  • Created demo-to-product-gap-march-of-nines (concept, status: active, conviction: medium-high): captures Karpathy's framework — each additional "9" of reliability is constant work; 5 years at Tesla advanced FSD by 2-3 nines; 2014 perfect-Waymo-demo to 2026 minimal-deployment as a second data point. Cross-tagged for thread/artificial-intelligence because the framework is load-bearing on Karpathy's broader AGI-timeline argument (see agi-timeline-decade-of-agents).
  • Updated tesla-fsd (entity): added Karpathy first-person operator framing under Strengths ("Tesla strategy more scalable" assessment) and under Weaknesses (2-3-nines-in-5-years data point with cross-link to the new mechanism page).
  • Updated waymo (entity): added a substantial "Operator framings from Karpathy" subsection with four sub-points — 2014 perfect-demo data point, "deployments still pretty minimal" + capex-uneconomical observation, teleoperation-in-the-loop claim (with explicit Karpathy's-own-caveat that he doesn't have the full extent), geofence-as-signal-quality-proxy speculation.
  • Updated index: added a new "People (cross-context)" subsection under Entities for andrej-karpathy; added demo-to-product-gap-march-of-nines under Concepts; added source entry under Sources. Annotated entity entries with "Updated 2026-05-15" markers so reviewers can see which pages picked up the Karpathy material.

Cross-context handoff: the artificial-intelligence ingest preceded this one and owns the Karpathy entity's physical home + the agi-timeline-decade-of-agents concept page. The career ingest follows.

Contradictions surfaced: 0 (Karpathy's teleoperation claim is flagged as needing corroboration but isn't a contradiction with existing wiki content — it's a new claim, marked unverified per his own caveat).

ingest2026-04-20 17:05
2026-04-20-autoresearch-tesla-fsd

First source for this thread. Autoresearch synthesis on Tesla FSD covering architecture (HW3/AI4/AI5, v13 transformer "world model"), NHTSA investigations (EA26002 on 3.2M vehicles for reduced-visibility failures, PE25012 on traffic violations), California CPUC regulatory stance (Tesla holds limousine/TCP permit, not AV permit), unsupervised robotaxi rollout (Austin/Dallas/Houston as of April 18, 2026), and comparison to Waymo + Mercedes Drive Pilot.

  • Created tesla-fsd (entity): primary entity for the thread; vision-only L2-supervised stack, remote-supervised robotaxi in 3 cities.
  • Created waymo (entity): sensor-fused fully driverless robotaxi; operational yardstick (~500k rides/week).
  • Created mercedes-drive-pilot (entity): only L3 consumer system in US; mapped-freeway ODD.
  • Created nhtsa (entity): US regulator; three concurrent Tesla FSD probes including EA26002.
  • Created vision-only-vs-sensor-fusion (concept): central architectural fork — Tesla alone vs everyone else.
  • Created fsd-safety-data-validity (concept): Koopman's critique of Tesla's public safety report methodology.
  • Created will-nhtsa-ea26002-trigger-fsd-recall (question): outcome of the 3.2M-vehicle Engineering Analysis.
  • Created is-vision-only-sufficient-for-l4-autonomy (question): the defining empirical question for the thread.
  • Created does-remote-supervised-robotaxi-qualify-as-driverless (question): terminology/regulatory line-drawing.
  • Updated index.md — replaced TODO summary, populated all four sections.
  • Contradiction flagged (unresolved): Musk's public "10× safer than humans" claim vs Koopman analysis + community-tracker regression on v14.2 — captured in fsd-safety-data-validity rather than silently reconciled.

note2026-04-20 16:50
thread created

Thread autonomous-driving scaffolded via /thread-create.

  • Scope: undefined — set in SCOPE.md when ready