The Memo - 5/Dec/2025
US Genesis Mission, 13 Neuralink participants, McKinsey reports, and much more!
To: US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients…
From: Dr Alan D. Thompson <LifeArchitect.ai>
Sent: 5/Dec/2025
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 95 ➜ 96%
ASI: 0/50 (no expected movement until post-AGI)OpenAI + universities (20/Nov/2025, PDF):
’...frontier models like GPT-5 are increasingly able to accelerate novel research… assisting in ideation, powering deep literature search, and even proposing and implementing proofs of (appropriately-sized) open problems. None of this would have been possible just twelve months ago. The fact that GPT-5 is already showing it can accelerate scientific research, combined with the rate of improvement of frontier AI, suggests that the coming years will be an exhilarating time in science and mathematics.’
Happy 3rd birthday to ChatGPT. How can it be the end of 2025 already? I’m definitely ready for 2026 and all it will bring. Full subscribers will receive my end-of-year ‘The sky is…’ report shortly…
My latest keynote, recorded in front of a live audience of 3,500 people in Sydney, is now available to full subscribers in Downloads & Highlights. We’re also coming up on The Memo roundtable #40, details at the end of this edition.
Contents
The BIG Stuff (Genesis Mission, 22 new models over the last 30 days…)
The Interesting Stuff (McKinsey, AI books preferred, Ive product, HF ecosystem…)
Policy (AI espionage, Europe scaling back AI Act, MIT, OpenAI, US jobs…)
Toys to Play With (ChatGPT use cases, M4 vs M5, FLUX.2, new AI font…)
Things I’ve Been Thinking About (AI relationships + a new story by Gemini 3 Pro…)
Next (Roundtable…)
The Memo features in recent AI papers by Microsoft and Apple, has been discussed on Joe Rogan’s podcast, and a trusted source says it is used by top brass at the White House. Across over 100 editions, The Memo continues to be the #1 AI advisory, informing 10,000+ full subscribers including RAND, Google, and Meta AI. Full subscribers have complete access to all 25+ AI analysis items in this edition!
The BIG Stuff
USA: Launching the Genesis Mission (24/Nov/2025)
The US President has launched the ‘Genesis Mission’, a groundbreaking initiative aimed at propelling the US to the forefront of artificial intelligence (AI) development. This mission is designed to emulate the urgency and ambition of the Manhattan Project, aggregating massive federal scientific datasets to train frontier models and create AI agents for rapid scientific advancement.
Sidenote: Here are the top initial collaborators sorted (very roughly) by market cap:
NVIDIA, Google/Alphabet, Microsoft, Amazon, AMD, Siemens, Cisco, Applied Materials, IBM, RTX (Raytheon Technologies), GE Aerospace, Synopsys, Dell Technologies, Micron, OpenAI for Government, HPE, Nokia, Albemarle, AVEVA (Schneider Electric), Anthropic, Scale AI, Cerebras, Hugging Face, Quantinuum, Sambanova…
Read more via The White House, and the official site.
I’ve provided an analysis of the likely datasets here: https://lifearchitect.ai/genesis/
This was added to the ASI countdown as an info point, and there have been several other significant indicators over the last 30 days: https://lifearchitect.ai/asi/
22 new models over the last 30 days (Dec/2025)
The last 30 days have been huge for model releases across labs in both the US and China. Here’s a summary of 22 model highlights from the Models Table:
Generalist GEN-0 (10B on 10T tokens)
Embodied foundation model built for multimodal training directly on high-fidelity raw physical interaction. (Playground, Paper)Moonshot AI Kimi K2 Thinking (1T on 15.5T tokens)
1T parameters and 384 experts. Open source SOTA. HLE=51.0 on text-only subset. (Playground, Paper)Google DeepMind HOPE (1.3B on 100B tokens)
Combining our self-modifying sequence model with the continuum memory system, we present a learning module called HOPE. (Paper)Baidu ERNIE-4.5-VL-28B-A3B-Thinking (28B on 15T tokens)
28B-A3B. Open-sourced 12/Nov/2025 from Jun/2025 release. (Playground, Paper)Tsinghua JustRL-Nemotron-1.5B (1.5B on 9T tokens)
JustRL, a simple recipe with fixed hyperparameters, achieves state-of-the-art performance on two different 1.5B base models while using 2× less compute. (Playground, Paper)OpenAI GPT-5.1
Personality change via fine-tuning. GPQA (no tools) increased from GPT-5=85.7 to GPT-5.1=88.1. (Playground, Paper)Baidu ERNIE-5.0-Preview-1022 (2.4T on 100T tokens)
Very low performance on ALPrompt. 2.4T params confirmed. (Playground, Paper)PleIAs Baguettotron (0.321B on 200B tokens)
The name is both a nod to French origins and to the unusual shape of the model: with 80 layers, Baguettotron is currently the deepest SLM in its size range. (Playground, Paper)xAI Grok 4.1 (3T on 80T tokens)
(Playground, Paper)Google DeepMind Gemini 3
The knowledge cutoff date for Gemini 3 Pro was January 2025. (Playground, Paper)Allen AI OLMo 3 (32B on 6T tokens)
Announce: https://allenai.org/blog/olmo3 (Playground, Paper)