The Memo - 13/Mar/2026
OpenAI GPT-5.4, China 70% AI target, data center construction surpasses office construction, 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: 13/Mar/2026
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 97%
ASI: 0/50 (no expected movement until post-AGI)A few hidden facts were revealed this month:
OpenAI CEO says ‘We are training right now on the first site in Abilene [Texas] what I think will be the best model in the world hopefully by a lot [likely the GPT-6 model].’ (BlackRock, 12/Mar/2026).
The US State Department has switched from Anthropic Claude to the older, non-reasoning OpenAI GPT-4.1 model (Reuters, 3/Mar/2026). That model was launched a year ago (Apr/2025), and was known for quality. It is interesting that they have not chosen any of the 19 other models1 in the newer GPT-5 series (yes, that’s 19 excluding GPT-5 Codex models!).
OpenAI has confirmed their latest headcount is 5,000 employees (VB, 3/Mar/2026).
Contents
The BIG Stuff (OpenAI GPT-5.4, China 70% AI target, DCs surpass offices…)
The Interesting Stuff (Huawei flatpack DCs, labor markets, Meta chips, AMI…)
Policy (US 92k job losses in Feb, AI Amicus brief, xAI datasets…)
Toys to Play With (Claude Viz, Share kindness, FinePhrase dataset, LLM book…)
Things I’ve Been Thinking About (Post-ASI games…)
Next (Roundtable…)
The BIG Stuff
China’s 141-page plan targets AI across 70% of economy within a few months (10/Mar/2026)
China’s 15th Five-Year Plan, unveiled at the National People’s Congress, places AI and robotics at the centre of its economic future, aiming for AI penetration across 70% of its economy by next year (2027) and 90% by 2030, with AI-related industries targeting a value exceeding US$1.38T. Analyst Shanaka Anslem Perera noted (10/Mar/2026): ‘This is not just an economic plan. It is a war plan for a war the United States is not fighting.’
Read analysis via Moneycontrol and Reuters.
A shorter briefing of the 15th Five-Year Plan is available (PDF, English, 47 pages), and the final Plan (141 pages) should be available in English later at: https://en.ndrc.gov.cn/policies/
AI takeover complete: data center construction surpasses office construction for the first time (28/Feb/2026)
The US Census Bureau’s Construction Spending report for December confirmed that the value of data centers constructed in the US has officially surpassed the value of offices for the first time, a historic crossover signaling that ‘going forward machines, and not human workers will provide the bulk of US productivity.’ AI demand is forecast to require US$3T in data center investment according to Morgan Stanley.
Read more via ZeroHedge.
OpenAI GPT-5.4 (5/Mar/2026)
OpenAI GPT-5.4 is the newest state-of-the-art frontier model. It matches or exceeds human industry professionals in 83% of comparisons on GDPval across 44 occupations, up from 70.9% with GPT-5.2. It is the first general-purpose model with native computer-use capabilities, supports 1M tokens of context, and minimizes hallucinations (33% less likely to contain false claims).
Read the announce, system card, try it at chat.com, see it on the Models Table.
The Interesting Stuff
TIME: How Anthropic became the most disruptive company in the world (11/Mar/2026)
In a 5,500-word article, TIME explores the inside story of Anthropic. Coverage includes the company’s confrontation with the Pentagon over autonomous weapons and mass surveillance red lines (leading to a first-of-its-kind supply-chain risk designation against an American company), consumer sign-ups surging past 1M/day, and Claude’s iPhone app displacing ChatGPT at #1. Anthropic’s alignment stress-testing lead Evan Hubinger frames the acceleration starkly: ‘Recursive self-improvement, in the broadest sense, is not a future phenomenon. It is a present phenomenon.’
Read more via TIME.
Google Home update lets Gemini describe live camera feeds (3/Mar/2026)
Google Home now features ‘Live Search,’ allowing Gemini to analyze and describe what your cameras see in real time, not just recorded events. Users can ask questions like ‘Hey Google, is there a car in the driveway?’ with the feature available on the Advanced plan of Google Home Premium at US$20/month or US$200/year.
Read more via The Verge, CNET, and the announce via Google.
Figure 03 in the living room (Mar/2026)
Today, Figure is showing another major milestone towards a robot in every home. Running Helix 02, cleaning a living room fully autonomously…
[confirmed not remotely operated, but] ‘fully autonomous’.
Figure AI’s Helix 02 system in the Figure 03 humanoid robot handles spray bottles, towels, pillows, remote controls, and bins through coordinated manipulation, in-hand reorientation, dynamic throws, and tight-space navigation, all learned purely from data. This is a single general-purpose architecture expanding its repertoire simply by adding more training examples, building toward ‘scalable humanoid intelligence’ for homes and workplaces.
Read more via Figure AI.
Watch the video (link, original source):
And advanced fine motor control skills are demonstrated in this close-up (source):
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!
LLMs can unmask pseudonymous users at scale with surprising accuracy (3/Mar/2026)
New research demonstrates that AI models can deanonymize pseudonymous social media users far more effectively than traditional methods, achieving up to 90% precision. Unlike older approaches, AI agents can start from free text like anonymized interview transcripts and work their way to a person’s full identity by browsing the web and using simulated reasoning.