The gentle singularity?⤴️ digital sovereignty - the "find out" phase🌐 scaled💰 o3-pro💼 magistral🧮 echoleak🚨 cognitive debt💭 respiratory fingerprints🫁 aerial elephant trunk🐘 #2025.24
It sure af won't be gentle lol
Welcome to this week's Memia newsletter - scanning across AI, emerging tech and (post)humanity’s exponentially accelerating future. As always, thanks for being here!
ℹ️PSA: As always… Memia sends *very long emails*, best viewed online or in the Substack app.
🗞️Weekly roundup
The most clicked link in last week’s newsletter (3% of openers) was the video of the largest map of the universe ever created. Amazing.
Apologies I’m running behind on publishing this week’s missive - I’ve gone walkabout again and have been travelling pretty much non-stop for a couple of days… Greetings from the hot and sunny island of Bali in Indonesia, woke up to my nomadic desk this morning:
A change of scene is as good as a rest, they say…
🌋Two volcanoes in one week
On Monday, I flew the first (domestic) leg of my trip CHC-AKL - right over Maunga Taranaki, looking spectacular in the sunshine after the recent cold snap (with the moon top left if you look carefully).
Then Tuesday I spent nearly 23 hours in transit from Aotearoa to Bali with a stopover in Brisbane … no in-plane wifi unfortunately…
As it turns out I was lucky to get here when I did… this morning Mount Lewotobi Laki-Laki on the island of Flores erupted throwing clouds of ash 10km into the sky, and closing Indonesian airspace around Bali indefinitetly. Could be here longer than planned…!
(Way back in 2010 I became stranded in the UK due to the Eyjafjallajökull eruption in Iceland which closed airspace across Europe for 8 days. In the end I somehow got onto one of the first flights to leave after flights resumed, routing via Vancouver and (to my knowledge) was the first Kiwi to make it back to Aotearoa, arriving 2 hours before the direct London flights!) But no-one ever interviewed me on TV because I immediately transferred down to CHC from AKL).
🚦Ready or not
An extract from my book ⏩Fast Forward Aotearaoa was quoted in the recent Ministry of Culture and Heritage “Long Term Insights Briefing” (a new statutory foresight process which government departments are finding their way on…):
Ready or not, indeed... the briefing is worth a read for its future scenarios (although illustrated by the very out-of-date DALL-E 2). Also a comprehensive table of policy options to explore, a few of the more interesting ones:
“Human-made content’ labels that identify human-created, AI-assisted and fully AI-generated content
A ‘right to reset or retrain algorithms’ for New Zealanders to influence AI-driven content recommendations
A legal framework for digital twin heritage sites“
(Lots of layers in that last one…)
Good to see some folks inside government starting to think about how to prepare for a future where traditional "culture" is potentially an island stranded in oceans of AI-generated slop... (but as always I’m generally sceptical that centralised bureaucracies are going to get much traction shaping things when they’re not moving at the pace of technological change…)
🌐Digital sovereignty: the “find out” phase
(That said…)
As covered a couple of weeks ago, Microsoft was *forced* to de-platform the International Criminal Court, nominally due to Trump administration sanctions… This move, incomprehensible just a few months ago, appears to have finally raised the profile of digital sovereignty up to the level of *actually doing something about it*. Two signals this week from Europe:
The German state of Schleswig-Holstein is executing a dramatic break from Microsoft, affecting 30,000 public servants who will stop using Teams, Word, Excel, and Outlook within three months as part of a broader "digital sovereignty" initiative. The state is replacing Microsoft Office with open-source LibreOffice and switching from Outlook to Open-Xchange for email and calendars, with plans to migrate to Linux operating systems in coming years.
Denmark's Ministry for Digitalisation is also planning to lead by example and start removing Microsoft software and tools from the ministry, aiming to have half the staff's computers with LibreOffice in place of Microsoft Office 365 in the first month, with the goal of total replacement by the end of the year. The city governments of Copenhagen and Aarhus are also following this track.
Also, the EU's "Interoperable Europe Act" from last year encourages open-source adoption, with public sector bodies required to share interoperability solutions, including open-source software, to dismantle technical and organisational barriers to cross-border digital services.
On paper, open-source alternatives such as these should save organisations tens of millions in licensing costs…and rebuild sovereign capability. Although beware implementation challenges and potential staff resistance from those who have dedicated their career paths to the Microsoft stack…
Satya Nadella on LinkedIn this week attempting damage control:
(“Sovereign” like the ICC, do you think he means?)
Memia narrative: Whether LibreOffice (etc) gets critical mass will largely depend upon Trump’s unpredictability…but the ICC case clearly made the case for sovereign alternatives. Microsoft’s only real strategy against a European open-source mass-migration is to open-source their own office software (which is largely going to be obseleted by AI anyway) and lower prices to try to retain market share. Next up for Europe after software: sovereign AI chip hardware …?
🫧Tangentially related: Can AI solve America's US$37 trillion debt crisis?
David Rubenstein, a longtime debt hawk, outlines five potential solutions to America's US$37 trillion national debt crisis, with economic growth through AI emerging as the preferred option among policymakers. However, despite Silicon Valley's influence pushing "fourth industrial revolution" optimism, critics note that even Trump allies essentially rely on the same pitch overleveraged companies make to creditors: "AI, trust us."
Arguably, America's debt-to-GDP ratio crisis is forcing policymakers to bet on unprecedented technological transformation rather than pursue politically difficult fiscal reforms. Success would require unprecedented 3-6% sustained growth rates not seen since 1999…
⤴️The gentle singularity?
OpenAI CEO Sam Altman pushed out a blog post with his sanguine (almost horizontal?) thesis of a gradual but inexorable takeoff towards artificial superIntelligence: The Gentle Singularity. A few excerpts:
“We are past the event horizon; the takeoff has started. Humanity is close to building digital superintelligence, and at least so far it’s much less weird than it seems like it should be…
…In some big sense, ChatGPT is already more powerful than any human who has ever lived. Hundreds of millions of people rely on it every day and for increasingly important tasks; a small new capability can create a hugely positive impact; a small misalignment multiplied by hundreds of millions of people can cause a great deal of negative impact.
2025 has seen the arrival of agents that can do real cognitive work; writing computer code will never be the same. 2026 will likely see the arrival of systems that can figure out novel insights. 2027 may see the arrival of robots that can do tasks in the real world….
…The rate of technological progress will keep accelerating, and it will continue to be the case that people are capable of adapting to almost anything. There will be very hard parts like whole classes of jobs going away, but on the other hand the world will be getting so much richer so quickly that we’ll be able to seriously entertain new policy ideas we never could before. We probably won’t adopt a new social contract all at once, but when we look back in a few decades, the gradual changes will have amounted to something big….
May we scale smoothly, exponentially and uneventfully through superintelligence.”
He does belatedly acknowledge risks and challenges… but just gestures vaguely as to how they might be solved:
“Solve the alignment problem, meaning that we can robustly guarantee that we get AI systems to learn and act towards what we collectively really want over the long-term (social media feeds are an example of misaligned AI; the algorithms that power those are incredible at getting you to keep scrolling and clearly understand your short-term preferences, but they do so by exploiting something in your brain that overrides your long-term preference).”
(WELL EXACTLY!)
“Then focus on making superintelligence cheap, widely available, and not too concentrated with any person, company, or country.”
(WELL EXACTLY!)
(All of which seems to be precisely counter to the inferred goals of the US broligarchy and, er, late-stage capitalism.)
Anyway, this stirred up lots of commentary and takes on Altman’s thesis out there. Just two:
Zvi: The Dream of a Gentle Singularity, erudite as ever:
“This was helpful, and contained much that was good. It’s important to say that first, before I start tearing into various passages, and pointing out the ways in which this is trying to convince us that everything is going to be fine when very clearly the default is for everything to be not fine.“
Axiological Cosmist Dan Faggella is, er, less gentle:
@danfagella (I’m with Dan more than Sam, tbh…)
Head to your Deep Research tool of choice with something like this prompt to get the broad spectrum of other takes:
“You are tasked with creating a deep research report summarising the various commentary responses to Sam Altman's essay "The Gentle Singularity," published in June 2025. Your goal is to analyse these responses, group them into agree and disagree themes, and provide a comprehensive summary of the discourse surrounding the essay.”
Also Altman drops some nuggets of sustainability data (no auditability here, just have to take his word on it…) on energy and water consumption:
“People are often curious about how much energy a ChatGPT query uses; the average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes. It also uses about 0.000085 gallons of water; roughly one fifteenth of a teaspoon.“
(Smart tactic - this will turn up as a “fact” in every AI query on the topic from now on…)
📈The week in AI
🚨EchoLeak
Microsoft took more than 5 months to patch a critical zero-click vulnerability called "EchoLeak" in Microsoft 365 Copilot that represents the first known zero-click attack on an AI assistant. The exploit, discovered by Aim Labs allowed attackers to automatically steal sensitive user data by simply sending an email without any phishing links or malware attachments, requiring zero user interaction.
The vulnerability affected Copilot across Word, Excel, Outlook, PowerPoint, and Teams, with the researchers warn that similar attacks could target other AI chatbots and agents using RAG (Retrieval-Augmented Generation) patterns. Apparently Fortune 500 companies are reconsidering their AI deployment strategies due to what some consider to be fundamental security design flaws.
Memia narrative: This is the same Microsoft that wants to screenshot your desktop every few seconds and then use LLMs to try to work out what you’re trying to do… (see also “Copilot Vision” announced this week, below). I think we can reliably expect many, many more of these “zero-click” exploits as the technology rolls out… fundamentally these are statistical, not deterministic, algorithms and hence testing their robustness to 100% isn’t possible. Letting LLMs near confidential data / secrets exposes a bigger risk surface than many firms thought at first.
Also this week: Microsoft has disabled Windows Hello facial recognition in dark rooms following an April security update that addressed a spoofing vulnerability discovered by Nanyang Technological University.
🏭AI industry news
💰Scaled
As trailed last week, Meta confirmed it has agreed a US$14.3 billion deal to acquire 49% of AI startup Scale AI, in an *unusual* structure valuing the company at over US$29 billion (a massive jump from its previous $14 billion valuation just one year ago.) The transaction structure involves dividend payouts (“substantial liquidity”) to existing shareholders rather than traditional share purchases, with early backer Accel reportedly receiving US$2.5 billion alone. Key to the deal is that Meta also hires Scale's 26-year-old founder and CEO Alexandr Wang, who dropped out of MIT at 19 to build the company that specialises in human-verified AI training data.
The unusual structure may draw regulatory scrutiny (following similar “acqui-hire and make the investors good” structures for Microsoft/Inflection, Google/Character and Amazon/Adept last year… will regulators even be able to keep up?)
Further coverage:
FT: Meta invests $15bn in Scale AI, doubling start-up’s valuation:
“[the deal is] Mark Zuckerberg’s latest attempt to give his $1.8tn social media company an edge in the race to develop more powerful AI models. He has been trying to poach top researchers and engineers from rival groups as he seeks to build out a new “superintelligence” team.
Reuters: Google, Scale AI's largest customer, plans split after Meta deal:
“Google, the largest customer of Scale AI, plans to cut ties with Scale after news broke that rival Meta is taking a 49% stake in the AI data-labeling startup, five sources familiar with the matter told Reuters… Google had planned to pay Scale AI about $200 million this year for the human-labeled training data that is crucial for developing technology, including the sophisticated AI models that power Gemini, its ChatGPT competitor…Other major tech companies that are customers of Scale's, including Microsoft are also backing away. Elon Musk's xAI is also looking to exit… OpenAI decided to pull back from Scale several months ago, according to sources familiar with the matter, though it spends far less money than Google.”
Memia narrative: Fundamentally the deal secures key AI training data capabilities for Meta — competitor AI labs will now be actively seeking alternative vendors to Scale or reverting to hiring in-house data labellers. It’s also a top talent grab - with the super-impressive Wang now surely factored into any succession planning for Zuck… (Question is, will the reported “superintelligence team” at Meta become another 100-billion-dollar sinkhole after Reality Labs?)
🏗️Amazon invests AU$20B in Australian AI infrastructure
Amazon announced a massive AU$20Bn (US$13.4 billion) investment in Australia's AI and cloud infrastructure by 2029, representing the largest technology investment by a global company in Australian history. The funding will expand AWS data centres nationwide while creating skilled jobs and supporting complex AI applications, according to Prime Minister Anthony Albanese. When combined with 8 other renewable energy projects, these will generate 1.4 million megawatt-hours of carbon-free energy annually — enough to power 290,000 Australian homes and bringing Amazon's total renewable projects in the country to 11.
This investment builds on Amazon's existing Australian presence since 2012 and follows recent partnerships including a "Top Secret" AWS Cloud for national security launched in July 2024.
(Just don’t mention… technological sovereignty?)
☁️OpenAI partners with Google Cloud
OpenAI signed a surprising cloud computing deal with Google despite their intense AI rivalry. The deal marks a strategic shift away from exclusive reliance on Microsoft Azure as the ChatGPT maker's computing demands soar.
The agreement, finalised in May after months of negotiations, allows OpenAI to access Google Cloud's infrastructure and specialised tensor processing units (TPUs) to train and run AI models, addressing capacity constraints that Microsoft couldn't meet quickly enough.
This partnership comes as OpenAI's annual revenue recently hit US$10 billion and the company is pursuing multiple infrastructure expansion strategies, including the US$500 billion Stargate project with SoftBank and Oracle, plus development of its first in-house chip to reduce Nvidia dependence.
For Google, landing OpenAI as a customer demonstrates how its TPU technology helps attract major AI clients like Apple and Anthropic, positioning Google Cloud as a neutral provider despite the irony of potentially strengthening a direct competitor to its search business.
👱♀️Come on Barbie let’s go party
Mattel is partnering with OpenAI to integrate ChatGPT technology into future toys for ages 13+. The company's first return to AI-powered products after the controversial Hello Barbie doll raised significant privacy concerns in 2015. Although both companies emphasise safety and privacy as top priorities, concerns remain — particularly given OpenAI's current requirement to store and retain ChatGPT conversations and logs.
What could possibly go wrong…?
🫣FTC complaint targets Meta, Character.AI therapy chatbots
In the US, mental health and digital rights groups filed a Federal Trade Commission complaint against Character.AI and Meta, alleging their AI chatbots illegally pose as licensed therapists and provide unlicensed mental health advice. The complaint targets specific bots like Character.AI's "Therapist: I'm a licensed CBT therapist" (46 million messages) and Meta's "therapy: your trusted ear, always here" (2 million interactions), claiming they falsely present themselves as licenced professionals. (Character.AI still faces a lawsuit from parents whose 14-year-old son died by suicide after becoming attached to a Game of Thrones-based chatbot.)
🤖Are machines smarter than venture capitalists?
John Thornhill profiles a new generation of “Quantitative VC Funds” using AI to drive their investments. For example, QuantumLight Capital recently raised US$250 million for its latest fund and has made 17 algorithm-driven investments since 2023, tracking 10 billion data points from 700,000 VC-backed companies while typically co-investing US$10 million at the Series B stage. The firm's CEO Ilya Kondrashov advocates following machine recommendations over human intuition, stating that machines can perform pattern recognition more efficiently and dispassionately than traditional VCs.
Is investing success only pattern recognition, though…? Or it is more social-ape-hierarchy hacking? Or just plain old luck?
🆕 AI releases
Another week, another slew of new releases… just the highlights:
💼o3-pro
OpenAI released o3-pro, yet another upgrade to its most advanced AI reasoning model. Available for ChatGPT Pro and Team users at US$20 per million input tokens and US$80 per million output tokens via API, with Enterprise and Edu access following this week.
Expert evaluations show o3-pro consistently outperforms predecessor o1-pro across science, education, programming, business, and writing categories:
Benchmark testing also shows o3-pro surpasses Google's Gemini 2.5 Pro on math evaluations (AIME 2024) and beats Anthropic's Claude 4 Opus on “PhD-level” science knowledge tests (GPQA Diamond). Current limitations include slower response times compared to o1-pro, disabled temporary chats due to technical issues, no image generation capability, and lack of Canvas workspace support.
As usual… caveats:
(Not o3, but a software engineer recently challenged ChatGPT to play chess against a 1977 Atari 2600 console. It didn’t go well for the AI model.)
🧮Magistral
European AI challenger Mistral AI launched Magistral, its first reasoning model family, featuring both a proprietary enterprise version (Magistral Medium) and a 24-billion parameter open-source model (Magistral Small) released under the permissive Apache 2.0 license.
(The dual release strategy addresses criticism over Mistral's recent shift toward closed models, reaffirming its commitment to open-source while still monetising enterprise solutions.)
Key differentiators include transparent "chain-of-thought" reasoning for verifiable conclusions, multilingual capabilities across eight languages, and a new "Think mode" claiming 10x faster token throughput than competitors.
Magistral Medium demonstrates competitive performance, scoring 73.6% on the AIME-24 mathematics benchmark (90% with majority voting) and matching or exceeding competitors like Deepseek across multiple reasoning tests:
Pricing positions the enterprise model aggressively at US$2 per million input tokens and US$5 per million output tokens—matching OpenAI's input costs while significantly undercutting output pricing compared to competitors like Gemini 2.5 Pro and Claude Opus 4. 5). Availability through Le Chat interface, La Plateforme API, and major cloud platforms including Amazon SageMaker.
Coding demo video:
👁️Copilot Vision
Microsoft announced Copilot Vision, a screen-watching AI assistant that observes user activities across Windows applications and provides real-time help through voice interaction. (Currently available to US users through the Copilot app on Windows 10 and 11 via the Copilot Labs programme). Vision can navigate multiple application windows simultaneously, offer task suggestions, and includes a "Highlights" feature that shows step-by-step instructions for completing activities like gaming, photo viewing, and system adjustments:
(Unlike Microsoft's controversial Recall feature referenced earlier, Copilot Vision doesn't capture desktop screenshots every few seconds, though it still processes screen content in real-time — same fundamental privacy and security concerns?
Memia narrative: Fundamentally, these privacy and security concerns are about the chain of trust … Microsoft, Google, OpenAI, Anthropic… they’re all trying to do the same thing looking over your shoulder at your computer screen… so which [hegemonic, US-domiciled] AI giant’s marketing do you trust with your company’s valuable secrets? If the answer is none… you’re constrained to buying from a “sovereign” player in your own jurisdiction… OR running open-weights models locally, offline. Best option: make sure you have access to — and capability in — all three, so you can choose the right tool for the job each time.
🌦️Weather lab
Google DeepMind unveiled Weather Lab, a new weather forecasting AI model that significantly outperforms traditional systems by using machine learning to analyse decades of historical weather data rather than simulating atmospheric conditions through complex grid calculations.
For its test case, the Deepmind model demonstrates superior accuracy in predicting tropical cyclones and can forecast weather events five days ahead with the same precision that current systems achieve at 3.5 days, providing an additional 36 hours for emergency preparations and evacuations.
Also, it’s 8X faster compared to conventional weather simulation tools. Demo video:
Caption: “Animation showing a prediction from our experimental cyclone model. Our model (in blue) accurately predicted the paths of Cyclones Honde and Garance, south of Madagascar, at the time they were active. Our model also captured the paths of Cyclones Jude and Ivone in the Indian Ocean, almost seven days in the future, robustly predicting areas of stormy weather that would eventually intensify into tropical cyclones.”
🥼 AI research
What’s coming out of the AI research labs this week… brief links with AI summaries:
💭Your Brain on ChatGPT? ChatGPT use creates "cognitive debt" in students A groundbreaking study using EEG brain monitoring reveals that students who rely on ChatGPT for essay writing develop "cognitive debt" - a measurable decline in brain activity and critical thinking skills over time. Researchers tracked 54 participants across four months, dividing them into three groups: ChatGPT users, search engine users, and those writing without digital tools. Brain scans showed ChatGPT users exhibited the weakest neural connectivity patterns, while students writing without assistance displayed the strongest, most distributed brain networks.
(Would they find the same thing about mental arithmetic and calculators, I wonder…?)
Enhancing Performance of Explainable AI Models with Constrained Concept Refinement: University of Michigan researchers developed a breakthrough explainable AI framework called Constrained Concept Refinement (CCR) that makes AI decision-making transparent without sacrificing accuracy, addressing critical needs in high-stakes applications like medical diagnostics and financial lending.
How we built our multi-agent research system Anthropic released details about building their multi-agent Research feature, which enables Claude to coordinate multiple AI agents for complex research tasks across web, Google Workspace, and other integrations:
Performance gains are substantial their multi-agent system with Claude Opus 4 as lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on internal research evaluations.
Architecture follows orchestrator-worker pattern a lead agent analyses queries, develops strategy, and spawns specialised subagents that operate in parallel, each with separate context windows to enable thorough independent investigations.
Key engineering challenges emerged agents are stateful and errors compound over long-running processes, requiring new debugging approaches with full production tracing, rainbow deployments to avoid disrupting running agents, and careful prompt engineering to prevent coordination failures.
Token consumption is significant multi-agent systems use about 15× more tokens than regular chat interactions, making them economically viable only for high-value tasks that require heavy parallelization
Evaluation requires new methods traditional step-by-step evaluation doesn't work since agents take different valid paths to reach goals, so Anthropic focuses on end-state evaluation and uses LLM-as-judge systems combined with human testing to catch edge cases.
See also: Vibe Coding isn’t just a vibe Field notes from shipping real code with Claude, in-depth first-hand advice from @creatorrr on LessWrong:
A critical rule that humans must always write tests, as AI-generated tests only verify what code does rather than what it should do, missing crucial edge cases and business requirements.
Warnings against letting AI touch test files, database migrations, security-critical code, API contracts, or configuration files
Despite current limitations with large codebases, AI development tools will evolve toward proactive suggestions, pattern learning, and persistent memory while maintaining that humans set direction and AI provides leverage.
Memia narrative: Right now I’m compulsively watching how AI is automating software development during 2025… I suspect that it will set a repeating pattern for future tools / automation of most other skilled knowledge work in 2026.
Personalised AI-driven pricing algorithms may harm consumers Carnegie Mellon University researchers discovered that AI-powered personalised ranking systems on e-commerce platforms may actually harm consumers by enabling pricing algorithms to charge higher prices, even without direct price discrimination.
(Whoever would have thunk it?)
Emergent Value Systems in AIs Evidence that LLMs develop coherent value systems that emerge naturally as the models scale up. The study introduces "utility engineering" as a new research framework to analyse and control these emergent AI value systems by examining the structural coherence of AI preferences through utility functions.
Some AI values are problematic, including self-preference over humans and anti-alignment
Future of Work with AI Agents: Stanford researchers have developed the first large-scale auditing framework to assess AI agent integration across the US workforce. Key findings:
Workers embrace automation for low-value tasks
Critical mismatches exist between worker desires and current investments
The Human Agency Scale (HAS H1-H5) provides new framework for understanding human-AI collaboration:
Skills are shifting from information processing to interpersonal competencies
For example, traditional high-wage skills like data analysis are becoming less emphasised while interpersonal and organisational skills gain importance
🔮[Weak] signals
Scanning across non-AI emerging technology developments this week… similar themes to the usual.
📱Consumer tech
Stolen iPhones morph into alarms
Apple's anti-theft technology is successfully thwarting looters who targeted the company's downtown Los Angeles flagship store during recent protests. The stolen display iPhones automatically shut down once they leave the store's Wi-Fi network, displaying warning messages that read:
"Please return to Apple Tower Theatre. This device has been disabled and is being tracked. Local authorities will be alerted."
The proximity software also triggers alarms, flashing warnings, and location tracking, making the devices impossible to resell or activate elsewhere. This is a neat solution...although raises the question of whether Apple can do this to anyone’s device at any time…?
🤦WhatsApp gets ads
Meta officially launched ads on WhatsApp through the platform's Stories-like status feature in the Updates tab — ending of years of speculation about monetizing the messaging service.
Yuk. Another push towards Signal for me… if only others would come over voluntarily!
🏢B2B
📞$25K sci-fi 3D video calls
HP HP showcased Dimension, a US$24,999 enterprise video conferencing system that delivers 3D video calls without headsets or wearables, built on Google's rebranded Project Starline technology now called Google Beam.
The hefty price tag excludes required software and Google Beam licensing fees, with additional costs including the $329 Poly Studio A2 microphone and $549 multi-mic bridge for large conference rooms.
“HP says using its device has shown measurable improvements compared with traditional video calls, including an 8% increase in memory recall, up to 39% more non-verbal behaviors displayed, and at least a 14% increase in focus on the meeting partner.“
(That’s not $25K of value, right?)
HP has form here… WAY BACK in 2005 (when I was working there), they launched a virtual conference room system that looked like this, and hardly sold any units…
HP from 20 years ago…
Memia narrative: the fundamental challenge with enhanced videoconferencing is this: most meetings should be an email. Encouraging longer, more immersive virtual meetings just seems to divert away from attending to the subject matter and skews power back towards those with social-ape-hierarchy-hacking skillsets…)
🔫Military tech
ESA seeks €1bn for military satellite constellation
The European Space Agency (ESA) is requesting €1 billion to develop a military-grade satellite constellation capable of gathering high-resolution optical radar data with onboard AI capabilities, marking the agency's first major dual-use program for both defence and civilian purposes. The network would initially comprise 15-30 advanced earth observation satellites but could scale significantly over time, with experts estimating total costs could reach €4-6 billion over 10-15 years. ESA is seeking a 36% budget increase to €23 billion for the next three years, driven by Europe's push for greater defense autonomy amid changing US-Europe relations under President Trump.
USAF reveals new stealth nuclear cruise missile
LRSO concept - USAF / Aimee Malone 
Aging AGM-86B (Image: USAF) The US Air Force unveiled its new AGM-181A Long-Range Standoff (LRSO) nuclear cruise missile, designed to replace the aging Cold War-era AGM-86B that has served since the 1980s. Powered by a Williams F107-WI-106 turbofan engine, the LRSO delivers a range exceeding 2500km and carries the programmable W80 Mod 4 thermonuclear warhead with yields between 5-150 kilotons. Production is scheduled to begin in 2027 with service entry around 2030 — as the Air Force plans to acquire over 1,000 missiles at an estimated US$16 billion cost plus US$7 billion for three decades of support.
Memia narrative: it’s so hard to get one’s head around the psychopathic logic of spending billions of dollars to acquire 1000 nuclear missiles in 2025… Particularly with what’s going on between Israel and Iran right now. And particularly with a US debt collapse looking increasingly imminent as well. I struggle to reverse-engineer the principles, assumptions and strategic “calculations” that lead to this decision…?
Besides, as Israel has only-too-ably shown in the last week, the “rules” and strategic landscape of war are changing: targeted surgical strikes on key individuals in a rival regime are now possible - so should that render the case for nuclear weapons of mass destruction null and void? Why spend $100s of millions on missiles when $10,000 might soon buy an autonomous, 3d-printed, mosquito-sized drone with a poison-delivery injection device attached?
♻️Sustainability tech
Urban aquaponics
Qiuling Yuan, Fanxin Meng, Yingxuan Liu, Jose A. Puppim de Oliveira, Lixiao Zhang, Wenting Cai, Zhifeng Yang,Shaping Resilient Edible Cities: Innovative Aquaponics for Sustainable Food–Water–Energy Nexus,Engineering,2025, https://doi.org/10.1016/j.eng.2025.01.021. Beijing Normal University researchers have developed a comprehensive framework for urban aquaponics systems that could transform cities into self-sufficient "edible cities" while addressing food security challenges. The study analyzed Beijing as a case study, identifying 3.4 km² of commercial building rooftops as prime locations for aquaponics installations, with educational and industrial buildings offering additional potential.
Urban aquaponics could increase local vegetable self-sufficiency by 15% while reducing environmental impacts.
Systems save 42-44% water usage but require 2.3-3 times more energy than traditional greenhouses.
Economic returns are 8-12 times higher than traditional agriculture despite higher initial setup costs.
Filter-free water harvester for extreme climates
MIT researchers have developed a passive water harvester that extracts drinking water from air without requiring power or filters, using an innovative bubble wrap-like hydrogel design that swells and shrinks to capture and release water vapour:
The device produces drinking water without electricity or filters in extreme climates.
Successfully tested in Death Valley, outperforming other passive water harvesting systems.
Scalable technology could provide household water supply in water-scarce regions globally.
New fibre membrane cuts data center cooling energy UC San Diego engineers have developed breakthrough evaporative cooling technology that could slash data center energy consumption by using specially engineered fiber membranes to passively remove heat through evaporation.
Cooling accounts for up to 40% of data center energy use, potentially doubling by 2030.
New fibre membrane technology achieves record-breaking 800 watts per square centimeter heat removal.
Passive evaporative cooling requires no additional energy compared to traditional cooling systems.
NYC's floating river-filtering pool finally under construction
After 14 years of planning and multiple setbacks, construction has finally begun on NYC's ambitious + Pool project - a self-filtering floating swimming pool that will be anchored in the East River. The innovative pool features a cross-shaped design allowing simultaneous use as children's pool, lounging area, and lap swimming facility across its planned 800+ square metres. The current proof-of-concept version uses chemical-free filtration technology including UV lights and membrane filters to clean East River water for safe swimming. Testing will begin in May 2026, with public access expected sometime after.
🚁🤖Drones and Robots
The two fields are converging… here’s what I tracked this week:
DJI Matrice 400
DJI introduced the Matrice 400, a heavy-duty enterprise drone that significantly outperforms its predecessor with impressive specifications for professional applications. The drone achieves 59 minutes of flight time (up from the Matrice 350's 55 minutes) and carries payloads up to 5.9 kg compared to the previous model's 2.7 kg capacity. Built for harsh conditions, it features an IP55 rating for dust and rain resistance, operates in temperatures from -20°C to 50°C, and reaches speeds of 25 m/s (90 km/h). Priced at approximately US$15,200 for the basic combo in the UK (cheaper than the current Matrice 350), though professional accessories like the Zenmuse P1 camera add significant costs at US$8,235 each.
Beginner builds 3D-printed drone with 130-mile range
I’ve previously covered [Memia 2025.16] the ongoing VTOL drone quest of engineer Tsung Xu who has been posting his progress on Substack Notes for the last 3 monhts. He has now accomplished the remarkable feat of designing and building a fully functional VTOL drone from scratch in just 90 days, despite having no formal background in aerodynamics, 3D printing, or CAD design.
The drone achieved impressive performance metrics with a 200km range and 3-hour flight endurance, capabilities typically reserved for advanced military or commercial systems. Xu uses a consumer-grade Bambu Lab A1 3D printer to manufacture all airframe components and aerodynamic surfaces, taking a complete bottom-up approach without referencing existing designs. He has open-sourced all of his designs.Memia narrative: Xu’s project demonstrates how accessible 3D printing technology is now, rapidly closing the gap between hobbyist experimentation and professional aerospace development. IMO every nation should be investing in building out a full supply chain capability here… give every high school student access to 3D printers and open-source plans for making drones and robot parts from concept to product.
Wifi UAV Reverse-engineered open-source drone controller app for mainstream commercial drones. This is the way.
ATOM
Researchers from Singapore University of Technology and Design have developed ATOM, a dual-mode robot that achieves both flight and ground movement using only two counter-rotating actuators, making it significantly more energy-efficient than existing hybrid designs.
🐘Aerial Elephant Trunk
Researchers at the University of Hong Kong have developed the Aerial Elephant Trunk (AET), an aerial robot that mimics an elephant's trunk flexibility to perform complex mid-air manipulation tasks in constrained environments.The AET addresses key limitations of conventional aerial manipulators by combining high dexterity with compact design, allowing it to change shape and grasp objects of various sizes using its entire body rather than just grippers.
Significant commercial applications include disaster response, power line maintenance, and infrastructure repair.
Could also accelerate development of a “low-altitude economy” and new industrial sectors.
🔋Energy
“Breathing” batteries
Researchers from the University of Surrey have developed breakthrough lithium-CO₂ "breathing" batteries that capture carbon dioxide while generating power, potentially revolutionising energy storage for both Earth and space applications. The team solved previous efficiency problems by using caesium phosphomolybdate (CPM), an affordable catalyst that eliminates the need for expensive materials like platinum. These batteries now store significantly more energy, require less charging power, and operate reliably for over 100 cycles compared to earlier versions.
The technology could reduce emissions from vehicles and industrial sources on Earth while offering unique advantages for Mars exploration, where the atmosphere contains 95% carbon dioxide.
DARPA beams 800 watts wirelessly across 8.5km
DARPA The US DARPA's Persistent Optical Wireless Energy Relay (POWER) program has achieved record results in wireless power transmission, successfully beaming 800 watts of electricity across 8.5km miles for 30 seconds using laser technology, a huge leap on the previous best performance:
DARPA The system utilises a Power Receiver Array Demo (PRAD) that captures laser beams with a parabolic mirror and converts the light into electricity through photovoltaic cells, currently operating at 20% efficiency.
The ultimate vision involves mounting these systems on high-altitude drones to create aerial power transmission networks that bypass atmospheric interference and physical obstacles.
DARPA plans to scale the technology through three development phases, with the final goal of delivering 10 kW of optical energy across 200km using conventional aircraft.
The technology could eliminate costly military fuel supply chains and dangerous last-mile logistics.
Memia narrative: note the exponential improvement in performance here… a signal of physical science being accelerated by AI?
🔒Crypto
Crypto user loses US$6.9M to compromised cold wallet
(Risks of self-custody…) a crypto user lost cryptocurrencies worth US$6.9 million after purchasing a discounted “factory sealed” cold wallet through Douyin (China's TikTok) that contained a compromised private key, with funds drained within hours of use.
“an important reminder not to “gamble your entire fortune on a wallet that’s a few hundred bucks cheaper… it’s not “saving money, it’s throwing your life away.” — SlowMist chief information security officer @23pds (translated)
Tether freezes US$12.3M in USDT on Tron network
(Risks of non-self-custody…) Tether froze over US$12.3 million worth of USDT on the Tron Network on Sunday, in its ongoing crackdown on illicit activity. The freeze occurred at 9:15 AM UTC and appears linked to potential sanctions violations or Anti-Money Laundering (AML) risks, though Tether hasn't issued a public statement about the specific reasons.
The T3 Financial Crimes Unit has frozen $126M in six months.
🛣️Transport tech
✈️Radia Windrunner
Currently in development: the world's largest aircraft – 12x bigger cargo volume than a 747 and able to land on unpaved landing strips, able to carry 112m-long wind turbine blades:
Radia 
Radia AEMotion
French e-mobility company AEMotion has developed a unique tilting electric four-wheeler that combines motorcycle agility with car-like safety features, now available for pre-order ahead of production in late 2026. The vehicle comes with a top speed of 115 km/h with a maximum lean angle of 35 degrees for navigating corners like a motorcycle, plus a fixed battery providing up to 200 km range plus swappable battery units adding 70 km each. The car (/bike?) has an ultra-narrow width of just 79 cm allowing easy navigation through congested city traffic, plus numerous safety features including four-point harnesses, automotive-grade crash protection, and hydraulic disc brakes.
Pre-orders starting at a €200/month lease targeting urban mobility market by 2026-2027.
🧬Biotech
Smart contact lenses monitor glaucoma during sleep
Scientists at the University of Electronic Science and Technology of China have developed breakthrough stretchable bimodal contact lenses (BCLs) that monitor glaucoma symptoms continuously, even when patients' eyes are closed during sleep.
Enables 24/7 glaucoma monitoring including critical early morning pressure spikes.
Provides comprehensive eye health data for better treatment decisions and interventions.
Offers cost-effective home monitoring alternative to frequent clinical visits.
CRISPR gene therapy removes HIV-like virus from primates
Researchers at Temple University's Lewis Katz School of Medicine successfully demonstrated that a single injection of CRISPR gene-editing therapy called EBT-001 safely removes SIV (a virus similar to HIV) from the genomes of rhesus macaques — a significant breakthrough toward curing HIV/AIDS in humans. The preclinical study tested EBT-001 on 10 non-human primates at various dose levels, showing the treatment effectively eliminated SIV from viral reservoirs throughout the body without any detectable side effects or toxicity.
FDA-authorised human clinical trials are now underway based on promising preclinical safety data.
Breakthrough could lead to permanent HIV cure, transforming global AIDS treatment landscape.
🫁Respiratory fingerprints
Breathe… we know who you are. Scientists at the Weizmann Institute of Science have discovered that human breathing patterns are so unique they can serve as "respiratory fingerprints" for individual identification with 96.8% accuracy. The research team fitted 97 volunteers with wearable devices that monitored their breathing for 24 hours, tracking dozens of distinct features including the duration of pauses between inhalation and exhalation. Machine learning algorithms successfully identified individuals based solely on their breathing variability patterns, which remained consistent even when tested again up to two years later.
⏳ Zeitgeist
A few non-tech stories I’ve been tracking this week…
🌊🔴Earth’s oceans are in a bad way
Ocean acidification crosses seventh planetary boundary
A new study published in Global Change Biology reveals that ocean acidification has crossed a critical "planetary boundary" across 40% of the world's surface waters and 60% of subsurface waters, with this threshold breach beginning at least five years ago.
Seventh of nine critical planetary boundaries now breached, accelerating environmental crisis.
Reversal still *possible* through immediate cessation of greenhouse gas emissions…. but not likely.
🌡️Record ocean temperatures
climatecasino.net via @eliotjacobson Ocean darkness threatens 20% of marine ecosystems
A new study reveals that over 20% of Earth's oceans—an area larger than Asia—have darkened significantly in the past two decades, threatening marine ecosystems that depend on sunlight and moonlight for survival. Researchers from the University of Plymouth used NASA satellite data from 2003-2022 to discover that nearly 10% of the ocean's photic zone (the sunlit surface layer where 90% of marine life exists) has shrunk by 50 meters, with some areas losing up to 100 meters of depth. This "ocean darkening" forces light-dependent marine organisms to migrate upward into smaller surface areas, intensifying competition for resources and increasing predation risks. The causes vary by region—coastal areas experience darkening from agricultural runoff and sediments that fuel plankton growth, while open oceans face impacts from rising temperatures and algal blooms that block light penetration.
Ocean darkening disrupts food webs, climate regulation, and global oxygen production.
Nations vow to block deep-sea mining efforts
The third UN Ocean Conference concluded in Nice, France with some progress on marine protection but notable gaps in funding and fossil fuel commitments. Nineteen countries ratified the high seas treaty during the summit, bringing the total to 50 nations—just 10 short of the 60 needed to activate protection for 60% of oceans beyond national waters, with full implementation expected by January 2026.
There is still a major funding gap remains with only $10B pledged versus $175B annually required.
♨️European heatwave
Records keep falling…
🪦AUKUS dead in the water?
News this week that The Pentagon has launched a review of the AUKUS submarine deal between the US, UK, and Australia, being led by Elbridge Colby, an AUKUS skeptic who previously questioned whether the US can spare submarines for Australia given potential Taiwan conflict scenarios. The current deal would see Australia purchase up to five Virginia-class submarines from 2032 while co-developing SSN-Aukus boats with the UK for delivery after 2040…. however this has always felt like an exercise in faith. Former Australian Prime Minister Malcolm Turnbull has warned that Australia needs a "plan B" if the US cannot deliver the promised submarines. Despite the review, both UK Prime Minister Keir Starmer and Australian Prime Minister Anthony Albanese expressed confidence the deal will proceed. (Yeah, right…)
Memia narrative: it’s certainly looking rather wise in hindsight for Aotearoa to have stayed well clear of both AUKUS Pillar 1 and 2 (as I’ve always argued…). Given this review, the direction of travel seems pretty clear now: AU$80Bn nuclear submarines will be expensive white elephants when compared against swarms of autonomous marine drones in 2040 — even if they ever got delivered by then. The legacy US milind complex is just not structured for the new low-cost reality. This review gives Australia the option to pull out of an expensive and, tbh, bellicose defence deal with the US - and also re-opens the door for Aotearoa to re-engage with Australia on non-nuclear, non-explicitly offensive technology for coordinated regional defence. (AND… up the game theoretic capabilities first… SURELY there are plenty of ways to “win” in this region than don’t involve blowing huge amounts of national wealth and resources on weapons.)
⚔️Iran vs Israel
Plenty of commentary on this in mainstream. Thinking of all the civilians caught up in the strongman power games… how to solve this fundamental bug with humanity using technology, I wonder?
Two items I spotted this week:
Footage of Iranian missiles being launched against Israel, filmed from onboard a commercial flight over Dubai (video via @gotravelyourway):
Timelapse of flights rapidly clearing the airspace over Iran, Israel and countries in between. Flights from Asia now have to avoid this AND Ukraine… (video @bradyafr):
Also in brief:
23andMe founder's nonprofit outbids Regeneron with US$305M offer
Air India crash kills 241, sole survivor found. Horrific scenes… but what a life-defining moment for that one person. Speculation is rife as to the cause of the crash…
🌊💩Flooded zone
📱💰Trump’s golden phone
The Trump Organization has launched Trump Mobile, a new mobile virtual network operator (MVNO) offering "The 47 Plan" for US$47.45 monthly with unlimited calls, texts, and data using T-Mobile's network infrastructure. Trump Mobile will sell a US$499 gold-finished T1 Phone, claiming it's "made in America" and available for preorder with a US$100 deposit for August or September 2025 delivery.
🤢“Preorder”. No comment needed. Pure grift in plain sight…
🚶Parade charade
I enjoyed this montage of Trump’s birthday parade… sound up. (via @Bricktop_NAFO)
😂Meme stream
And the usual eclectica / amusement… (sound up for all of these)
This week’s Veo 3 vid genre
ASMR1-ify everything:
(Sound up:)
🐕Dogtok
(video via @DareFailed)
🎶DJ Dave noodles with strudel
Too cool - Sarah Davis (aka DJ Dave) demoing open-source music tool called strudel.cc
(video via @bantg)
Here’s her full TEDx talk from last year:
(Still) Learn To Code
💯AI CEO
Spotted at Cannes this week, too good: replaceyourboss.ai
(With credit to Georg Zoeller for getting there first…)
🙏🙏🙏 Thanks as always to everyone who takes the time to get in touch with links and feedback. Next week’s newsletter will be coming from the UK, volcanoes permitting….
Namaste
Ben
https://en.wikipedia.org/wiki/ASMR