Sources & credits

Source deck snapshot: 17 September 2026. Restored and revised at Greg’s direction on 30 September 2026. Selected research updates are dated below; other source claims retain their original scope.

Baker v Macquarie University [2026] FWC 3054
Kieran Snyder, 9 April 2024
Anthropic Economic Index: initial report (10 February 2025)
METR Time Horizons 1.1 · 8 May 2026 snapshot
METR: Measuring AI Ability to Complete Long Tasks (2025)
Blender Python API documentation
Nathan Lambert, GPT-4.5 size estimate (28 February 2025)
DeepSeek V4 technical report: 1.6T total parameters
Epoch AI model data: Grok 4.7, Musk-stated 2.1T (checked 30 September 2026)
scaling01, speculative GPT-6 Astra estimate (28 September 2026)
GPT-3: 175B parameters (2020)
PaLM: 540B parameters (2022)
SemiAnalysis, reported GPT-4 architecture (10 July 2023)
OpenAI, GPT-1 architecture (2018)
OpenAI, GPT-2 model sizes (2019)
Scott Alexander · 2 July 2026
Forecasting Research Institute · 16 July 2026
Greg Baker: Let's quantify how unevenly spread the future is (13 July 2026; checked 30 September 2026)
GPT-4 · 2023-03-14
GPT-4 Turbo · 2023-11-06
GPT-4o · 2024-05-13
GPT-4.1 · 2025-04-14
GPT-5 · 2025-08-07
Claude 3 Opus · 2024-03-04
Claude Opus 4.5 · 2025-11-24
DeepSeek V3.2-Exp · 2025-09-29
GPT-6 Astra · 2026-09-30
Claude Fable 5.1 · 2026-09-30
DeepSeek V4 Pro 0813 · 2026-09-30
Anton Korinek, Economic Policy Challenges for the Age of AI (2024)
Berkeley Lab, Queued Up: 2026 Edition (US generation connections)
IEA, Global Critical Minerals Outlook 2025
OpenAI July incident update
OpenAI August incident report
Australian Government, 24 September 2026
NotebookLM is now Gemini Notebook
ElevenLabs receptionist

Presentation and artwork

This edition follows Greg Baker’s Pilot-01 deck with his 30 September corrections, selected TEC v6 slides, the retained forecasting slide and a new security-incident slide. The earlier AI Futures title is restored; the 49th Floor journey and US–China–Japan slide are omitted for this presentation. API prices, incident evidence and tools lists have dated updates. The Generative AI Trends slide retains the TEC v6 two-chart layout and human-brain synapse comparison. Its parameter chart includes selected historical counts and September 2026 updates, distinguishing disclosed counts from estimates and vendor claims. The original training-data chart and illustrative 2025 projections are retained. The private host introduction and General Assembly branding are omitted.

The METR chart replots all 26 results in the latest public snapshot, dated 8 May 2026 and checked 30 September. It shows 50% success estimates and uncertainty for predominantly software, ML and cyber tasks. Estimates above 16 hours are unreliable; several newer models remain unevaluated. All charts link to their primary sources.

Narration

Selected slide narration is adapted from Greg’s 1 October 2026 session, with room logistics and participant details removed. Other slides retain their earlier speaker notes or slide text. Audio is available for those twelve adapted slides and the retained forecasting slide. It is AI-generated using Greg’s existing voice, rather than a recording of the keynote.

Your information

Course progress and workbook answers are stored in your browser. The optional presenter voice controls require a private access code. Microphone speech is sent to OpenAI for transcription; completed speech is retained privately on merah with its timestamp and slide, and can be downloaded after signing in. External games have their own interfaces and data handling.