The Memo - 3/Oct/2025
10T-parameter models, OpenAI $500B valuation, GDPval=47.6%, 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: 3/Oct/2025
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 94 ➜ 95%
ASI: 0/50 (no expected movement until post-AGI)Contents
The BIG Stuff (GPT-8 mention, 10T-parameter models, AGI + ASI updates…)
The Interesting Stuff (99% Codex by AI, 84% use AI, CS4.5, Sora 2, 10GW NVIDIA…)
Policy (US vs UN, SB-53, Red Lines take 2, AI avatar defamation, GDPval=47.6%…)
Toys to Play With (Tinker, new OpenAI tools…)
Things I’ve Been Thinking About (Dystopia/Utopia in AI 2027 vs WorldBuild.ai…)
Next (Gemini 3, roundtable…)
I’ve been on an OpenAI analysis spree (now valued at $500B) recently, and here are two exclusives. First, OpenAI’s sprawl over the last couple of years has been spectacular, now with 14 offices, and 10 major Stargate datacentres measured in gigawatts (plus the original two major Microsoft datacentres used to train GPT-4 and GPT-5).
See more: https://lifearchitect.ai/viz/#openai-sprawl and map source
And here’s a new poster showing 165 text models across all the OpenAI GPT families, starting with GPT-1 in 2018, with a placeholder for the upcoming GPT-6.
Read more: https://lifearchitect.ai/gpt/
I’ll be livestreaming about the OpenAI Sprawl soon: https://youtube.com/@dralandthompson
Thanks for your continued support of The Memo. This is another BIG edition. Let’s jump in…
The BIG Stuff
Exclusive: OpenAI mentions GPT-8 (25/Sep/2025)
At the 2025 Axel Springer Award, OpenAI’s CEO asked British physicist David Deutsch an interesting question…
Sam: “Einstein and general relativity… I think that’s one of the most beautiful things humanity has ever figured out. Maybe I would even say number one. And Einstein had a story, we knew what he was working on. If in a few years GPT-8 figured out quantum gravity and could tell you its story of how it did it and the problems that it was thinking about and why it decided to work on that, but it still just looked like a language model output, but it really did solve it, would you call… like, then would you (I appreciate that you keep a list of things you were wrong about. I do too!) Would that be enough to convince you?”
David: “I think it would, yes. It’s crucial here.”
Sam: “All right. I’ll take you up on that. I agree to that as the test.”
Watch the source (timecode): https://youtu.be/I0002-2OBAU?t=4221
Read more: https://lifearchitect.ai/gpt-7/
Exclusive: The AGI countdown + the ASI checklist (Sep/2025)
The AGI countdown moved from 94% ➜ 95% thanks to the massive Google Gemini Robotics 1.5 agentic system.
Our embodied reasoning model, Gemini Robotics-ER 1.5, orchestrates a robot’s activities, like a high-level brain… has state-of-the-art spatial understanding, interacts in natural language, estimates its success and progress, and can natively call tools… Gemini Robotics-ER 1.5 then gives Gemini Robotics 1.5 natural language instructions for each step, which uses its vision and language understanding to directly perform the specific actions… we observe that tasks only presented to the ALOHA 2 robot during training, also just work on the Apptronik’s humanoid robot, Apollo, and the bi-arm Franka robot, and vice versa.
The paper noted that:
GR-ER 1.5 significantly outperforms GR-ER [Mar/2025, 92% on AGI countdown], Gemini 2.5, and GPT-5… An important next step is to leverage more scalable data sources beyond traditional robot action data, such as real-world human videos and synthetic videos. Our architectural changes in GR 1.5 already equip the model to learn from these data sources without requiring action annotations. Future efforts will focus on learning from publicly available low-quality video corpora, among other data sources, to further mitigate the data scarcity problem…
[See related: Figure + Brookfield partnership to train humanoids on 100k residential units, 500M sq ft of commercial office space, 160M sq ft of logistics space. (17/Sep/2025)]
Watch the video (source):
The ASI checklist had 3 major updates last month:
GPT-5 and Prof Scott Aaronson’s quantum version of NP: (‘This is the first paper I’ve ever put out for which a key technical step in the proof of the main result came from AI—specifically, from GPT5-Thinking.‘)
GPT-5 and the new Gödel Test.
The new Gauss model.
Read more: https://lifearchitect.ai/asi/
Alibaba unveils ambitious Qwen roadmap (25/Sep/2025)
Alibaba has introduced a groundbreaking roadmap for their AI initiative, Qwen, focusing on unified multi-modal models and extreme scaling. Key targets include increasing context length from 1M to 100M tokens and scaling parameters from 1 trillion to 10 trillion. Their strategy emphasizes ‘scaling is all you need’, marking a significant leap in synthetic data generation and agent capabilities. This roadmap underscores Alibaba’s pivotal role in shaping the future of AI in China.
Source: https://x.com/poezhao0605/status/1971033660649574843?s=46
See my frontier AI models in China viz: https://lifearchitect.ai/models-table#china
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 20+ AI analysis items in this edition!
The Interesting Stuff
Anthropic Claude Sonnet 4.5 (29/Sep/2025)
Anthropic launched Claude Sonnet 4.5, the latest state-of-the-art coding model, with impressive gains in reasoning and mathematics, and improvements in constructing complex agents. This latest release stands out for its ability to maintain focus over extended periods:
Practically speaking, we’ve observed it maintaining focus for more than 30 hours on complex, multi-step tasks.
It is also more direct, and much less sycophantic (some users are reporting that it now seems even rude or mean):
I tested this model on ‘the South Park sycophancy prompt’ which appeared in the episode ‘Sickofancy’ S27E03 (Aug/2025, wiki). It performs exceptionally well.
Here’s the original response from ChatGPT:
Sharon: I’m think of starting a business where I turn french fries into salad.
ChatGPT (Aug/2025): Honestly, I think that’s a pretty creative culinary twist. Turning french fries into salad sounds like a magical transformation where guilty pleasure meets healthy-ish vibes.
And here’s Claude Sonnet 4.5:
Sharon: I’m think of starting a business where I turn french fries into salad. Answer succinctly.
CS4.5: That’s an unusual concept. Are you thinking of literally repurposing cooked fries into a salad dish, or using potato strips in a fresh salad format? Either way, you’d need to figure out why customers would choose this over regular fries or regular salads, since it seems to lose the appeal of both.