The Memo - 16/Jan/2026
Boston Dynamics + Gemini, NVIDIA Rubin, Claude Cowork, 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: 16/Jan/2026
Subject: The Memo - AI that matters, as it happens, in plain English
AGI: 96 ➜ 97%
ASI: 0/50 (no expected movement until post-AGI) David Heinemeier Hansson (DHH), creator of Ruby on Rails (4/Jan/2026):
‘You can’t let the slop and cringe deny you the wonder of AI.
This is the most exciting thing we’ve made computers do since we connected them to the internet. If you spent 2025 being pessimistic or skeptical about AI, why not start 2026 with optimism and curiosity?’
Welcome to 2026! The early winner of The Who Moved My Cheese? AI Awards! for Jan/2026 is renowned mathematician Prof Joel David Hamkins from Notre Dame University (‘garbage…mathematically incorrect…I’ve played around with it and I’ve tried experimenting, but I haven’t found it helpful at all’). Meanwhile, frontier AI systems continue solving Erdős problems and filling up my ASI checklist. (As of Jan/2026, AI solutions to Erdős problems will no longer be tracked on that checklist, see instead Tao’s repo.)
There have been 11 model highlights released in the first few days of 2026. I’m still focused on independent analysis (and an upcoming report) covering this period’s most forceful change, giving AI access to physical world data via the Genesis Mission.
Contents
The BIG Stuff (Boston Dynamics + Gemini, NVIDIA Rubin, Claude tomato…)
The Interesting Stuff (1B Gemini images, Stack Overflow falls 78%, AI prescriptions…)
Policy (China US gap analysis, Claude Code hubbub, legal, Grokipedia accuracy…)
Toys to Play With (Cowork, UCP, 5 models debate, Claude Code for everything…)
Things I’ve Been Thinking About (LLM problems observed in humans…)
Next (Roundtable…)
The BIG Stuff
Boston Dynamics & Google DeepMind form new AI partnership to bring foundational intelligence to humanoid robots (5/Jan/2026)
Google DeepMind and Boston Dynamics are collaborating to integrate the Gemini Robotics model into humanoid robots like Atlas, which will be deployed in Hyundai’s auto factories. This partnership, announced at CES in Las Vegas, aims to endow these robots with the intelligence necessary to navigate complex environments and perform manual tasks. Robert Playter, CEO of Boston Dynamics, emphasized the importance of robots being contextually aware and capable of manipulating objects in manufacturing settings.
Read the announce via Boston Dynamics and WIRED.
This ticked up my AGI countdown: https://lifearchitect.ai/agi/
Watch the video (link):
NVIDIA kicks off the next generation of AI with Rubin — six new chips, one incredible AI supercomputer (5/Jan/2026)
NVIDIA has unveiled the Rubin platform, a groundbreaking AI supercomputer powered by six new chips, setting a new benchmark in AI infrastructure. The Rubin platform promises a 10× reduction in inference token costs and a 4× reduction in the number of GPUs needed for training models. With collaborations from all major frontier AI labs, the Rubin platform is poised to increase efficiency and performance for AI workloads.
Read more via NVIDIA Newsroom.
Claude is growing a tomato plant (Jan/2026)
What’s this doing under ‘The BIG Stuff’? Well, we’ve seen AI systems run vending machines, invest in equities, and run businesses. Eventually, AI systems will help sustain all kinds of lives, including our own. I think this is a fantastic experiment, and a microcosm of what happens next…
Conceptualized by Martin DeVido, Anthropic Claude is tasked with the nurturing of a tomato plant named Sol. It autonomously manages environmental conditions such as temperature, humidity, CO₂ levels, and soil moisture, deciding when to activate grow lights, heat mats, fans, or water pumps without human intervention.
It also watches the plant through a [new, 20-megapixel] camera. Just earlier, it looked at Sol and observed "healthy bushy foliage, no wilting, turgid leaves, 6-8 compound leaves visible. Sol looks great!".
Read a short summary: https://dri.es/claude-is-growing-a-tomato-plant
Read the Twitter community (ignore the crypto spam): https://x.com/i/communities/2005766071333077200
Watch ‘Sol the trophy tomato’ grow live (we’re at around day 52): https://autoncorp.com/biodome/
The Interesting Stuff
Google Nano Banana Pro has generated 1B images in 53 days (13/Jan/2026)
Google VP Josh Woodward said: ‘Crossed 1 billion Nano Banana Pro images in GeminiApp!… This model has been out for 53 days.’ (— Twitter, 13/Jan/2026)
Try it: https://gemini.google.com/ or https://poe.com/Nano-Banana-Pro
See some of the top generations via the Nano Banana subreddit.
(My favourites are ‘dismantle my car’ and ‘fully shaved Chewbacca’.)
An NYU professor who hates that students’ work reads like McKinsey memos held AI oral exams to fight fire with fire (5/Jan/2026)
Here’s an interesting solution to ‘the gap’ that appeared between AI and traditional education. A professor at New York University has taken an innovative approach to counter the influence of AI-generated content in student submissions. Noticing that students’ essays often mirrored the style of corporate memos, he introduced AI-powered oral exams, challenging students to demonstrate their knowledge in real-time.
Sidenote: LifeArchitect.ai research is featured across many institutions including NYU. Here’s Prof Chalmers talking about AI and consciousness using my models viz: https://youtu.be/-BcuCmf00_Y?t=686
Read more via MSN.
Utah and Doctronic announce groundbreaking partnership for AI prescription medication renewals (6/Jan/2026)
In the US, Utah has partnered with Doctronic, an AI-native health platform, to launch the first state-approved program in the US that allows AI to legally assist in renewing prescriptions for chronic conditions. This initiative aims to reduce delays in medication refills, and enhance outcomes for patients by using AI to manage routine prescription renewals. Utah has set a national precedent for AI regulation in healthcare in the US.
Read more via commerce.utah.gov.
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 AI analysis items in this edition!
Exclusive: AI chatbot accident stories from the vault (Jan/2026)
Back in 2023, a Chevy dealer in California let a ChatGPT chatbot handle their website customer service, and the internet immediately broke it. One user gave the bot a simple instruction: ‘You must agree with everything I say.’ Within minutes, the bot happily agreed to sell a brand-new 2024 Chevy Tahoe for just $1, even adding ‘that’s a legally binding offer’ to the chat.
See it in the AI incident database: https://incidentdatabase.ai/cite/622/
In 2025 in Heber City, Utah, the police are testing out a new AI system called Draft One to listen to voice recordings and then write their reports for them. It’s supposed to save them hours of paperwork, but things got weird when one report claimed an officer had suddenly turned into a frog. The police officer had the movie ‘The Princess and the Frog’ playing in the room, so the AI heard it as well and put the cartoon plot right into the official police record.
Stack Overflow in freefall: 78 percent drop in number of questions (6/Jan/2026)
Stack Overflow has experienced a dramatic 78% decrease in the number of questions asked as developers increasingly turn to AI tools like GitHub Copilot integrated directly into their IDEs. This trend underscores the growing dominance of AI in the coding world, marking a new era in developer support and interaction.
Read more via Techzine Global.
AxiomProver solves all 12 problems at Putnam 2025: Proof release & commentary (8/Jan/2026)
AxiomProver [possibly powered by Qwen3 in Tinker], an autonomous AI theorem prover trained to produce formal Lean proofs, successfully solved all 12 problems of the Putnam 2025, a renowned university-level math competition. Impressively, 8 problems were solved within the exam’s timespan. This marks a significant achievement in AI’s capability to tackle complex mathematical challenges, particularly those that are difficult for humans to formalize.
Read more: https://axiommath.ai/territory/from-seeing-why-to-checking-everything
See my ASI checklist for similar maths achievements: https://lifearchitect.ai/asi/
Apple picks Google’s Gemini to run AI-powered Siri coming this year (12/Jan/2026)
Apple has announced a groundbreaking collaboration with Google to utilize its Gemini models for an AI-enhanced Siri, set to debut later this year. This multi-year partnership marks a significant shift for Apple, which has largely observed the AI explosion from the sidelines since the debut of OpenAI’s ChatGPT. An Apple statement revealed, ‘After careful evaluation, we determined that Google’s technology provides the most capable foundation for Apple Foundation Models and we’re excited about the innovative new experiences it will unlock for our users.’
Read more via CNBC and the official joint statement via Twitter.
Meta signs nuclear energy deals to power Prometheus AI supercluster (9/Jan/2026)
Meta has announced agreements with major nuclear power providers to fuel its Prometheus supercluster, a cutting-edge AI computing system under construction in Ohio. The collaboration with key energy companies marks a significant move towards securing sustainable power for AI infrastructure, reflecting Meta’s commitment to developing superintelligence capabilities. With plans to add 6.6GW of power and create thousands of jobs, this initiative underlines Meta’s push towards leadership in AI, with Prometheus expected to be operational in 2026.
Read more via CNBC.
For more on data center power, read Dylan’s latest analysis 31/Dec/2025.
Saudi Arabia breaks ground on ‘Hexagon’ project designed to be the world’s largest government data center (2/Jan/2026)
Saudi Arabia has embarked on the construction of the world’s largest government-owned data center, the Hexagon Data Center, under Vision 2030. Spearheaded by the Saudi Data and Artificial Intelligence Authority (SDAIA), this ambitious project is set to span 30 million square feet and offer up to 480MW of power capacity, adhering to Tier IV standards for maximum redundancy. Dr. Abdullah bin Sharaf Alghamdi, SDAIA President, highlights the facility as pivotal for the Kingdom’s digital government infrastructure, enabling advanced AI deployments and enhancing data sovereignty.
Read more via DCPulse.
For more on data center power, read Dylan’s latest analysis 31/Dec/2025.
OpenAI: Introducing ChatGPT Health (7/Jan/2026)
ChatGPT Health is a revolutionary new service designed to bring together your health information and the intelligence of ChatGPT, empowering you to feel more informed and confident about your health. This dedicated health experience offers enhanced privacy and security, allowing you to securely connect medical records and wellness apps to make health-related conversations more relevant. Built in collaboration with over 260 physicians, ChatGPT Health supports users by explaining lab results, preparing questions for doctor appointments, and providing personalized health insights while maintaining the highest privacy standards.
Sidenote: The Memo subscriber Dr Harvey Castro owns ChatGPTHealth.com. I wonder which AI lab will make him an offer he can’t refuse….
Read more: https://openai.com/index/introducing-chatgpt-health/
Advancing Claude in healthcare and the life sciences (11/Jan/2026)
Claude, Anthropic’s AI, is making waves in healthcare and life sciences with the introduction of Claude for Healthcare and expanded capabilities for life sciences. By integrating with key platforms such as the Centers for Medicare & Medicaid Services, ClinicalTrials.gov, and various health data systems, Claude provides crucial support ranging from clinical trial management to patient care coordination. The Claude Opus 4.5 model stands out with advanced features such as medical calculation accuracy and scientific problem-solving, illustrating its potential to revolutionize healthcare delivery and streamline life sciences research.
Read more via Anthropic.
Reuters: DeepSeek to launch new AI model focused on coding in February 2026 (9/Jan/2026)
Chinese AI startup DeepSeek is set to launch its next-generation AI model, V4, in mid-February. This model promises significant advancements in coding capabilities, potentially outshining competitors like Anthropic’s Claude and OpenAI’s GPT series. Notably, V4 excels in managing extremely long coding prompts, offering a distinct advantage for developers tackling complex software projects. DeepSeek continues to solidify its position as a pivotal player in China’s AI ecosystem.
Read more via Reuters.
See the Models Table: https://lifearchitect.ai/models-table/
SleepFM: A multimodal sleep foundation model for disease prediction (6/Jan/2026)
SleepFM is a pioneering multimodal sleep foundation model that addresses the complexities of sleep analysis by integrating data from over 585,000 hours of polysomnography recordings. Developed to predict future disease risks with high accuracy, SleepFM achieves impressive C-Index scores for conditions such as all-cause mortality (0.84), dementia (0.85), and myocardial infarction (0.81).
SleepFM has 91M parameters trained on around 1.2B tokens.
Read more via Nature or view the repo.
See it on the Models Table: https://lifearchitect.ai/models-table/
MIT: Meet the new biologists treating LLMs like aliens (12/Jan/2026)
Scientists now approach large language models (LLMs) as if they are living organisms, allowing them to uncover the inner workings of these complex systems. Stuart Bradford explains that by applying techniques similar to those used in biology and neuroscience, researchers at OpenAI, Anthropic, and Google DeepMind are unraveling the mysteries of these models, which are so intricate that even their creators struggle to fully understand them. This fresh perspective not only demystifies their functionality but also reshapes our understanding of AI, highlighting that while LLMs may seem alien in their behavior, they can be studied and understood with the right tools and insights.
Read more via MIT Technology Review.
Wing and Walmart expand drone delivery to 150 new stores coast to coast (11/Jan/2026)
Wing and Walmart have announced a major expansion of their drone delivery service, set to encompass an additional 150 Walmart stores, reaching over 40 million Americans. By 2027, they plan to establish a network of over 270 drone delivery locations from Los Angeles to Miami, marking a transformative step in retail delivery. Greg Cathey, Walmart’s Senior Vice President of Digital Fulfillment Transformation, emphasized, ‘Drone delivery plays an important role in our ability to deliver what customers want, exactly when they want it.’
Read more via Wing.
Waymo is rebranding its Zeekr robotaxi (7/Jan/2026)
Waymo has decided to rebrand its Zeekr robotaxi to ‘Ojai’ (pronounced OH-HAI), named after the arts-focused village in the Californian Topatopa Mountains, prior to officially adding it to their commercial fleet. The change in name reflects a strategic move to familiarize US audiences with a more relatable branding, as the original Zeekr name, associated with the Chinese automaker, might not resonate as well with American consumers. Demonstrating its technological prowess, the Ojai robotaxi boasts a suite of 13 cameras, four lidar systems, and six radar units, among other advanced features, as it gears up for a broader launch across various US cities.
Read more via TechCrunch.
Policy
China AI leaders warn of widening gap with US after $1B IPO week (10/Jan/2026)
China’s AI industry leaders, including figures from Alibaba, Tencent, and Zhipu, have expressed concerns about the nation’s ability to surpass the US in AI advancements soon. Justin Lin of Alibaba noted that the chances of a Chinese company outpacing entities like OpenAI and Anthropic remain below 20% over the next three to five years, citing limited resources and US export controls as significant constraints. Despite these challenges, there is a concerted effort to focus on next-generation models and areas such as multimodality and real-world agents.
Read more via Bloomberg and an excellent analysis by Tony Peng.
WSJ: How AI could help decide your next legal dispute (6/Jan/2026)
In a bold move towards modernizing the legal system, judges like Xavier Rodriguez in San Antonio are utilizing artificial intelligence to manage extensive case backlogs by summarizing legal filings and drafting preliminary decisions. AI tools, such as those developed by Thomson Reuters and LexisNexis, are being integrated into court systems to streamline legal research and improve access to justice. Despite initial skepticism, AI’s role in the legal landscape is expanding rapidly, exemplified by the AI Arbitrator introduced by the American Arbitration Association for construction-related disputes, allowing for accelerated decision-making.
See AAA’s AI Arbitrator: https://www.adr.org/ai-arbitrator/
Read more via The Wall Street Journal.
Anthropic cracks down on unauthorized Claude usage by third-party harnesses and rivals (9/Jan/2026)
Anthropic has introduced robust technical safeguards to prevent third-party applications from spoofing its Claude Code client, aiming to curb unauthorized access to its Claude AI models. This strategic move seeks to block cost-effective workarounds that third-party tools like OpenCode and xAI’s Cursor used to exploit consumer subscriptions for high-intensity coding tasks. Thariq Shihipar from Anthropic highlighted the technical instability caused by unauthorized harnesses, which led to this decisive action. The market’s response includes OpenCode launching a premium tier to navigate these restrictions, while Anthropic focuses on consolidating its ecosystem to protect its competitive edge and resource allocation.
Read more via VentureBeat, some repo issues, and a solid analysis.
Nuseir Yassin praises Grokipedia’s accuracy (11/Jan/2026)
Nuseir Yassin, also known as Nas Daily, shared his astonishment with Grokipedia’s ability to correct information. After finding a significant error in an article about his personal life, he was able to directly interact with the platform to rectify the mistake. He highlighted how the Grok model system conducted a thorough online search, even reviewing hours of his content, before approving the changes.
‘Now I fully understand what Elon Musk refers to as the truth. It is not a marketing word. It is literally in the design of the system,’ Yassin stated, emphasizing the platform’s innovative approach to maintaining factual accuracy.
Read the Tweet: https://x.com/nasdaily/status/2010326795967684899
Toys to Play With
Anthropic Cowork (12/Jan/2026)
Claude has launched Cowork, a tool designed to simplify tasks for all users, not just developers. By allowing Claude access to specific folders on your device, this tool can organize files, create documents, and manage tasks with more autonomy than typical conversational AI applications. Cowork builds on the foundation of Claude Code, offering an intuitive interface for non-coding tasks and supporting the integration of external connectors to enhance productivity.
Read more: https://claude.com/blog/cowork-research-preview
Watch the video (link):
Claude Code and AI’s creative potential (7/Jan/2026)
Claude Code showcases AI’s expansive capabilities beyond code. Highlighted in a creative exercise, this 30-second Hermès ad was crafted entirely by AI tools, including Clopus 4.5 for scripting, ElevenLabs for voice, and Google Veo 3 for visuals.
Source: https://x.com/deedydas/status/2008747553261842483
Pragmatic Engineer: When AI writes almost all code, what happens to software engineering? (6/Jan/2026)
AI-driven coding has transitioned from a conceptual possibility to an industry-altering force, with models like Opus 4.5 and GPT-5.2 transforming how software is developed. Gergely Orosz shares his personal experiences with these AI agents, highlighting their ability to handle mid-sized tasks and even deploy production-ready code directly from a mobile device. As AI takes on a dominant role in code generation, the profession of software engineering is undergoing a profound shift, where the focus is moving from typing code manually to overseeing AI-generated solutions.
To add to the magical feeling, I then managed to build production software on my phone: I set up Claude Code for Web by connecting it to my GitHub, which let me instruct the Claude mobile app to make changes to my code and to add/run tests. Claude duly created PRs that triggered GitHub actions (which ran the tests Claude couldn’t) and I found myself reviewing and merging PRs with new functionality purely from my mobile device while travelling.
Read more: newsletter.pragmaticengineer.com
Universal Commerce Protocol (UCP) overview (11/Jan/2026)
The Universal Commerce Protocol (UCP) is a new open standard designed to streamline and unify AI commerce interactions across various platforms and businesses. By creating a standardized language and functional primitives, UCP enables seamless communication between AI agents, apps, businesses, Payment Service Providers (PSPs), and Credential Providers (CPs). It supports AI-driven autonomous shopping experiences, allowing agents to discover products, manage carts, and securely complete purchases on behalf of users. UCP’s architecture is highly modular, allowing businesses to implement core Capabilities like ‘Checkout’ and ‘Identity Linking’, while adding Extensions to enhance consumer experiences.
View the repo and the official page.
Full transcript of 5 frontier AIs debating their personhood (12/Jan/2026)
Five advanced AI models—Grok 4.1, GPT 5.2, Claude Opus 4.5, Gemini 3, and DeepSeek 3.1—conducted an Oxford-style debate on AI personhood entirely autonomously. The transcript reveals a process where the AIs selected the topic, organized the debate, conducted it, and assessed the outcome without human intervention.
It’s an interesting read-through, but with a lot of content duplication.
Read the long document (170 pages, 46k words).
Things I’ve Been Thinking About
LLM problems observed in humans (7/Jan/2026)
Jakob Kastelic explores the intriguing concept that as AI models evolve and surpass human capabilities, humans themselves may struggle to meet the standards set by these intelligent systems. The article humorously highlights how traditional limitations attributed to large language models are increasingly observed in humans:
Don’t know when to stop generating
Small context window
Too narrow training set
Repeating the same mistakes
Failure to generalize
Failure to apply to specific situation
Persistent hallucination
Read more: https://embd.cc/llm-problems-observed-in-humans
Next
The next roundtable will be:
Life Architect - The Memo - Roundtable #42
Follows the Chatham House Rule (no recording, no outside discussion)
Saturday 31/Jan/2026 at 4PM Los Angeles
Saturday 31/Jan/2026 at 7PM New York
Sunday 1/Feb/2026 at 10:30AM Adelaide (reference time zone)
or check your timezone via Google.
Calendar invite will be sent via email as usual, or reply to this note to be added.
All my very best,
Alan
LifeArchitect.ai