GPT 4.5🤷 comet incoming☄️ sesame🗣️ wetware-as-a-service🧠 chirp🐦 evoflow-rna🧩 Xpandomer🧬 FLORA💮 stellaris☀️ blue ghost👻 RIP skype🪦 #2025.09
Pressure cooks diamonds💎
Welcome to the Memia’s weekly scan across emerging tech and the exponentially accelerating future. Another full one this week, strap in…
ℹ️PSA: Memia sends *very long emails*, best viewed online or in the Substack app.
🗞️Weekly roundup
The most clicked link in last week’s newsletter (2% of openers) was the video of the first stratospheric flight of the Kea Atmos Mk1b — reaching 56,000 feet altitude (you can see my back garden from up there…!)
🎤Upcoming speaking
I’m going to be back on the speaker circuit around Aotearoa in the next month, a few dates for your diary below, hope to see you there:
Fri 21 March Taking part in a panel discussion The Transformative Power of AI at the AI Health Innovation Hub event 21st March at Auckland City Hospital.
Wed 26th March AUT Technology in Society Series event: AI for Aotearoa at AUT City Campus, Auckland:
A conversation between AUT Vice-Chancellor, Professor Damon Salesa and AI futurist and author Ben Reid discussing:
AI in 2025 – the latest technologies and what’s next?
Will AI take our jobs? Or create thousands of new roles?
What does AI mean for the future of education?
How should Aotearoa adapt to maximise AI benefits?
(Free entry, registration required)
Thu 10th April Keynote at the Moananui Out Of The Blue Summit in Nelson:
How Exponential AI, Robotics and Data are shaping the future of New Zealand’s Blue Economy.
(Reminder: reach out to book me for the most up-to-date AI keynote for your upcoming conference or in-house event here).
✨Cosmic
Firstly: looking out to the cosmos… (far more interesting than obsessing over the amygdala of an attention-seeking, orange-tinted gerontocrat on the other side of the world…)
Physicists have been amazed by the observation of a super-energetic neutrino zipping through an enormous underwater detector located 3,450 meters deep in the Mediterranean Sea near Sicily.
In a paper published last month in Nature, researchers from the KM3NeT (Cubic Kilometre Neutrino Telescope) collaboration reported evidence from February 2023 of the highest-energy cosmic neutrino ever detected — a nearly massless particle with an estimated energy of about 220 PeV (220 x 1015 electron volts or 220 million billion electron volts) — 10,000 times more than that generated in the largest particle accelerator on Earth. The origin of this neutrino remains unclear: the scientists peered out into space and identified 12 distant galaxies as possible origination points, though the neutrino may have arisen from some other source. (Indeed it may be cosmogenic - from the Big Bang itself…)
Anyway, a new window opened up into the universe's most extreme phenomena such as supermassive black holes and gamma-ray bursts.Humanity should think a lot more about this than petty wars over national borders on our tiny planet, don’t you think?
(🎩 to my bro Joe who sent this my way.)
⚡📊Energy economics
Also published in nature this week: The momentum of the solar energy transition. From the abstract:
“Decarbonisation plans across the globe require zero-carbon energy sources to be widely deployed by 2050 or 2060. Solar energy is the most widely available energy resource on Earth, and its economic attractiveness is improving fast in a cycle of increasing investments. Here we use data-driven conditional technology and economic forecasting modelling to establish which zero carbon power sources could become dominant worldwide. We find that, due to technological trajectories set in motion by past policy, a global irreversible solar tipping point may have passed where solar energy gradually comes to dominate global electricity markets, without any further climate policies. Uncertainties arise, however, over grid stability in a renewables-dominated power system, the availability of sufficient finance in underdeveloped economies, the capacity of supply chains and political resistance from regions that lose employment. Policies resolving these barriers may be more effective than price instruments to accelerate the transition to clean energy.“
Lots of interesting graphs and charts accompanying the paper, but this one stands out the clearest:
Of course, the challenge today in 2025 is incumbents who are unwilling to watch their fossil-fuel balance sheets be reduced to zero. A reminder that despite plummeting renewables costs, most of the world’s energy supply is still fossil-fuel derived:
Given forecasts for huge AI-driven energy demand growth in the next few decades, we may well see Jevon’s Paradox at play here - even though the economics of solar+storage makes sense for all new investments, the growth in overall energy demand and reluctance to give up oil and gas revenues will likely sustain fossil fuel energy sources for decades to come?
At what cost to the planet’s climate? How to change this?
📈The week in AI
A more subdued week than usual, except of course the release of GPT 4.5 (see below). It’s almost as if *other events* have been capturing everyone’s attention to the exclusion of all other narratives…?!
🏭AI industry news
Nvidia's quarterly earnings report kept the *exuberant* AI market afloat for a little bit longer - annual revenue soaring 114% YoY to a staggering US$130.5 billion as AI chip demand continued to grow. (But for how much longer? As I’ve asked before, who is Nvidia’s customer’s customer…?)
May not last. As of yesterday, NVDA shares were down significantly to US$113…
…dragging along most of the rest of the US market, and then the S&P 500 erased a whopping $1.5 trillion in market cap in one day after Trump’s tariffs announcement. (See Flooded Zone below).
The Economist assesses the likelihood of continued US export restrictions on Nvidia’s GPUs under the Trump administration: Nvidia is fighting both Trump and China.
Also reports this week of arrests in Singapore related to grey-market GPU sales to China … finger in the dyke, even if Trump does extend the Biden restrictions…?
Anthropic got their timing just right, raising another US$3.5Bn at a US$61.5 billion valuation, led by Lightspeed Venture Partners.
Google co-founder Sergey Brin sparked controversy by advocating for 60-hour in-office workweeks as the "sweet spot of productivity" for Google's AI teams. (Thanks but, er, no thanks Grandpa).
Wayve The UK AI start-up is rapidly expanding globally after securing over US$1bn in funding from major investors including SoftBank, Microsoft, and Nvidia — testing of its driving autonomy systems has begun in Germany and US with potential large automaker deals in the pipeline.
📉Economics of AI, exhibit 1: flagged last week, Microsoft has dramatically scaled back its data center expansion plans. According to Ed Zitron, cancelling leases and letting Letters of Intent expire totalling over 1 gigawatt of capacity—equivalent to 14% of its current infrastructure.
Zitron’s take is worth reading in full, he pulls no punches on the dynamics at play here:
“Here's a question: if generative AI had so much demand, why is Microsoft canceling data center contracts?
… Microsoft's "shift in appetite for capacity [was] tied to OpenAI," which heavily suggests that Microsoft is, at best, less invested in the future of OpenAI, a statement confirmed by The Information, which adds that Microsoft had been trying to "lessen its reliance on OpenAI technology as they increasingly compete in selling AI products."
At worst, this situation could suggest that Microsoft is actively dumping OpenAI, and is having questions about the fundamentals of the generative AI industry.
In very plain terms, Microsoft, despite its excitement around artificial intelligence and its dogged insistence that it’s the future of technology, has canceled data center leases and over a gigawatt of other datacenter infrastructure. Doing so heavily suggests that Microsoft does not intend to expand its data center operations, or at least, to the extraordinary levels the company initially promised.“
I’m still expecting that market correction, any day now…
📈Economics of AI, exhibit 2: DeepSeek completed its week of open source releases… finishing up with this pearl of a Tweet:
…which set the X AI commentariat on fire:
So basically: DeepSeek is somehow operating the world’s most profitable Generative AI company — all running on the previous generation of hardware because they aren’t allowed the latest generation of Nvidia GPUs — based almost entirely on the team’s engineering prowess. As is a frequent techbro quip on X:
“pressure cooks diamonds“
General response to the DeepSeek open-source releases has been "wow, these guys are super-smart”.
Oh, and another DeepSeek innovation this week: Nighttime API price cuts of up to 75% to optimise resource utilisation during off-peak hours 00:30 to 08:30 Beijing time.
According to a CNBC report, Meta is planning to debut a standalone AI app during the second quarter this year to compete with ChatGPT (and Claude and DeepSeek and Google and Apple and xAI and Mistral and all the others…chat apps are a bit meh these days eh?)
SamA onto it as always:
🆕 AI releases
Only one real AI release story this week:
🤷GPT-4.5 After months of impatient waiting, OpenAI *finally* launched GPT-4.5 last Thursday — the company’s “largest and best model for chat yet.“
“By scaling unsupervised learning, GPT‑4.5 improves its ability to recognize patterns, draw connections, and generate creative insights without reasoning.
Early testing shows that interacting with GPT‑4.5 feels more natural. Its broader knowledge base, improved ability to follow user intent, and greater “EQ” make it useful for tasks like improving writing, programming, and solving practical problems. We also expect it to hallucinate less.”
As per the roadmap indications previously given by CEO Sam Altman, GPT4.5 is not by itself a multi-step reasoning model, but is the linguistic bedrock upon which GPT-5 (when it arrives) will be built:
“We advance AI capabilities by scaling two paradigms: unsupervised learning and chain-of-thought reasoning. Scaling chain-of-thought reasoning teaches models to think before they respond, allowing them to tackle complex STEM or logic problems. In contrast, scaling unsupervised learning increases world model accuracy, decreases hallucination rates, and improves associative thinking. GPT-4.5 is our next step in scaling the unsupervised learning paradigm.“
(Cast your mind back to September last year and the long-hyped-up release of OpenAI’s o1 “Strawberry” reasoning model, showing these two axes of scaling:)
Some key benchmark differences between GPT 4.5, GPT-4o (language model only) and o3-mini (OpenAI’s current frontier CoT reasoning model) are shown below:
(The SWE-Lancer “benchmark” is based on a new research paper from OpenAI: SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?) Yikes😬 - this goes right to the heart of OpenAI’s objective function
Full GPT-4.5 system card here.
Crowdsourced AI benchmarking platform LMArena also confirmed that GPT 4.5 tops all the main benchmark categories (although see below Karpathy’s questioning whether benchmarking is really valid any more, lmarena can be gamed these days…):
Currently GPT 4.5 is only available as a “research preview” - and if you’re not paying your US$200/month for the ChatGPT Pro subscription then… you’ll have to wait, OpenAI are out of GPUs, and StarGate is a while away yet…:
And if you want to use the API… that’s another way to max your credit card out:
“GPT‑4.5 is a very large and compute-intensive model, making it more expensive than and not a replacement for GPT‑4o. Because of this, we’re evaluating whether to continue serving it in the API long-term as we balance supporting current capabilities with building future models.”
Commentary has been somewhat muted:
Oops:
Nathan Lambert: GPT-4.5: "Not a frontier model"?
“OpenAI's latest model raises more questions than answers, but no, the AI bubble isn't popping quite yet.“
(The “not a frontier model” narrative is apparently largely down to public AI safety commitments which OpenAI have previously made… which one would guess in the current political climate are being tacitly de-emphasised…)
Another interesting nugget here is his calculation of what “biggest model yet” actually means:“Estimates place GPT-4.5 as about an order of magnitude more compute than GPT-4. These are not based on any released numbers, but given a combination of a bigger dataset and parameters (5X parameters + 2X dataset size = 10X compute), the model could be in in the ballpark of 5-7T parameters total, which if it had a similar sparsity factor to GPT-4 would be ~600B active parameters.“
Ethan Mollick:
On the same topic, Andre Karpathy thinks that there’s an AI model evaluation crisis - many new frontier models are overfitted to max out benchmark scores — but their true differentiation increasingly becomes a matter of “taste” for users. How do you create “benchmarks” for these more subtle manifestations of [super]intelligence?
(All these links and many more in Zvi Mowshowitz - On GPT-4.5 )
GPT 4.5 has got sharp edges, alright:
🤷So far a big non-event for most of us until it turns up in the $20/month ChatGPT Plus sub. Bring on GPT-5 so we can see what OpenAI’s billions really buy (or not).
Other notable AI releases this week:
☄️Comet incoming Not strictly a release, but challenger Perplexity upped the ante on Google again, teasing a new “browser for agentic search” coming soon:
(Sign up for the waitlist here: https://www.perplexity.ai/comet)
Are browsers about to become a crowded space once again? The Browser Company (makers of the short-lived Arc Browser project) teased their own “AI Browser” Dia at the end of last year.
Meanwhile, over at Firefox: Mozilla recently faced significant backlash due to changes in its privacy policies and terms of use. This git diff says it all:
(Mozilla have since walked back the wording of the proposed privacy changes… but the legalese still feels a bit cack-handed…thankfully with open source there isn’t anywhere to hide…)
Phi-4 Microsoft announced Phi-4-multimodal and Phi-4-mini, the newest models in the Phi family of small language models (SLMs):
“Phi-4-multimodal, with its ability to process speech, vision, and text simultaneously, opens new possibilities for creating innovative and context-aware applications. Phi-4-mini, on the other hand, excels in text-based tasks, providing high accuracy and scalability in a compact form.“
Released under an open licence via HuggingFace, the models benchmark well against other small models including Gemini-2-Flash-lite-preview and Claude-3.5-Sonnet. More grist to the mill that Microsoft is rapidly reducing its single-point-of-failure dependency on OpenAI.
Dragon Copilot also from Microsoft: Microsoft Dragon Copilot, an AI assistant for clinical workflow:
“the healthcare industry’s first unified voice AI assistant that enables clinicians to streamline clinical documentation, surface information and automate tasks”
Basically: automated AI dictation and note-taking for doctors, nurses and other health professionals. Should help save a lot of time… although the Dragon^Copilot branding…yikes.
Mercury All leading LLMs to date have been built on the transformer architecture, whereas most generative AI image models tend to use diffusion architectures (here’s the difference). Now Inception Labs has released Mercury:
“the first commercial-grade diffusion large language model (dLLM). dLLMs push the frontier of intelligence and speed with parallel, coarse-to-fine text generation.”
🗣️Sesame released a preview of their state-of-the-art expressive voice technology. Only West Coast US accents so far, but impressive nonetheless:
(Check out the two release preview voices Miles and Maya having a chat to each other…). Sesame’s website trails some kind of smart glasses device:
“Voice as an interface is nuanced and intimate, which makes it a difficult medium to get right. There is a high bar that must be met for voice interfaces to be useful and not annoying, and we are careful and deliberate in how we design our products to meet our users where they want to be.
We believe that a great voice product calls for an interdisciplinary approach that tightly integrates hardware, software, and machine learning.“
…Definitely a new player to watch.
🙉”Accent-softening” Also on the subject of voice…Teleperformance SE, the world's largest call centre operator, is implementing “accent-softening” AI technology that “softens” Indian and Filipino call centre workers' accents in real-time to improve customer understanding... potentially countering the threat of chatbots replacing human agents. Potentially. (That’s going to be a weird thing to get used to if it catches on…and think about from an ethics POV…)
🐦Chirp when AI needs to talk directly to AI … better to Chirp than speak in English.
Each character is mapped to a frequency, so it can communicate to another client via chirps:
(Similar concept to the open source ggwave Gibberlink library:)
💮FLORA This is very cool. Flora, an intelligent canvas for every creative AI tool, connected.
🥼AI Research
Some good oil this week:
MASt3R-SLAM Real-Time Dense SLAM with 3D Reconstruction Priors… one of the best-performing models so far recreate an accurate 3D model from 2D video. Here’s an example of a recreation of the prison from The Shawshank Redemption: (via @bilawalsidhu):
AI societies in Minecraft startup Altera, led by Robert Yang, formerly an assistant professor at the MIT Department of Brain and Cognitive Sciences,
launched Project Sid (after Sid Meier, the creator of the iconic Civilisation video game). They instantiated 1,000 autonomous AI agents into a Minecraft world, and sat back to watch them build a society. Demo video below, and David Mattin looks deeper into the philosophical significance: Simulating the Post-Human Future.
🧩EvoFlow-RNA a masked diffusion model with the ability to unconditionally generate (and optimise) novel, naturalistic non-coding RNAs. The research team claims SOTA results on RNA representation learning, unconditional RNA design, and more.
(Shared by this guy, talking here about Google’s forthcoming AI Co-Scientist covered briefly last week:)
Seems legit.
🔮[Weak] signals
Non-AI tech signals from near and far futures...
📱Consumer tech
🪦RIP Skype Microsoft announced that Skype will be shutting down in May 2025. I remember being among the earliest users of the European video calling app using it when it was dominant in the mid 2000s… but everything went downhill after it was acquired by Microsoft in 2011 for US$8.5 billion. “Teams me” doesn’t have the same ring to it, eh…?
Tecno Mobile unveiled the Spark Slim concept smartphone, boasting an ultra-thin profile of just 5.75mm at its thinnest point—significantly slimmer than Apple's iPhone 16 (7.8mm) and even thinner than a standard pencil (7mm). Just don’t sit down with it in your back pocket.
Lenovo unveiled its Yoga Solar PC concept at MWC 2025, featuring solar charging capabilities that can provide one hour of video playback after just 20 minutes in direct sunlight.
Framework the open-source modular (repairable!) laptop leader announced three major products at its '2nd Gen' event: A modular desktop PC featuring AMD Ryzen AI Max processors in a 4.5L Mini-ITX form factor, the new Ryzen AI 300 Series Framework Laptop 13 and a touchscreen Laptop 12 which converts to a tablet. All pre-loaded with Linux of your choice (or Windows if you want…)
Framework In an era of increasingly crunchy supply chains… I can feel a tailwind behind open-source, modular, repairable hardware like this.
🔐Crypto
Last week I mentioned that the global Cardano community had ratified a democratic constitution on their blockchain. Here’s a rather uplifting documentary on how they did it:
(Shout out to Jo Allum, Engie Tumaru Matene and Robert O’Brien representing Aotearoa New Zealand so ably.)
🧫Biotech
🧬Xpandomer Pharma giant Roche released details of its new biotechnology Xpandomer, part of its Sequencing by Expansion (SBX) approach, a novel method in DNA sequencing that significantly enhances the accuracy and throughput of nanopore sequencing. Wow, real-world nanotech is here.
🦾Robots
Leapting's AI-powered PV Module Installation Robot is revolutionising solar panel installation in Australia, with the first units recently shipped to a 350 MW solar farm in New South Wales. The robot installs 60 modules per hour, reducing solar installation costs by 30% through automation.
Unitree continue to upgrade the Unitree G1's algorithm, enabling it to learn and perform virtually any movement. For example: Kung Fu (but no bow at the end?)
Sanctuary AI robots now come equipped with sensitive touch sensors:
Apptronik has launched a pilot partnership with manufacturing firm Jabil to test its humanoid Apollo robots in factory settings, with the eventual goal of having these robots build more of themselves by 2026. (Self-replicating humanoid robots, what could possibly go wrong…?)
Bio-inspired jumping robot
Harvard Harvard researchers have developed a bio-inspired jumping robot that mimics the springtail, a tiny hexapod capable of impressive leaps up to 23 times its body length.
📦Logistics
Automated mega port
Singapore's Tuas mega port, set to become the world's largest automated facility, has successfully handled 10 million containers since operations began in September 2022 and aims to process 65 million containers annually by the 2040s. The automated port uses digital twin technology to enable remote operation with minimal human workers on-site, and reduces carbon emissions by 50% through electrification and smart grid management.
🏗️Construction
The Wave House data centre in Heidelberg, Germany, constructed in October 2023, is the largest 3D printed building in Europe. It took only 140 hours, (less than six days) to build. (video via @guillaume_acc)
⚡Energy
Two very different nuclear energy startups came out of stealth this week:
☀️Proxima Fusion scientists at European startup Proxima Fusion published a peer-reviewed paper outlining Stellaris, a new stellarator-based concept for a nuclear fusion plant, leveraging advances in high-temperature superconducting magnets and computational optimisation. The company isplanning to build its first demonstration stellarator, Alpha—which “will show net fusion energy in a concept capable of continuous operation for the first time” in 2031.
⚛️Valar Atomics in the US, founder Isiaiah Taylor announced his new nuclear startup (yet another US tech bro leveraging Tolkien for naming inspiration… )
“Valar is my master plan to make energy 10x cheaper in 10 years by pulling oil and gas out of thin air with nuclear fission. This will untether energy from climate and politics, fuel American industry, and unlock a new era of growth…
…The atmosphere contains all the ingredients you need to make oil and gas. There are actually only two: hydrogen (from water) and carbon (from CO2). But you need a cheap power source to extract those elements and make them into oil and gas. That power source is nuclear fission“
Right…imagine one of these in every town…
The Valar team have been busy building the first prototype, keep an eye but the economics are…. suspect in my mind.
🚄Transport
We were promised…
Flying car Alef Aeronautics released footage of what CEO Jim Dukhovny claims is "the first documented, verifiable flight of a flying car" with vertical takeoff and no tethers. Crazy… apparently the company has taken pre-orders worth nearly $1 billion despite the US$300,000 price point.
Electric Helicopter Here in Aotearoa, inventor Oskar Stielau, based in Auckland, has been working on building a fully electric battery-powered helicopter since 2019. This test video from 2021…(Typical kiwi demo… no helmet or safety equipment, just shorts and a t-shirt…🫣) I imagine things have come on significantly since… the next model will be completely carbon fibre?
Electric seagliders Abu Dhabi's Strategic Development Fund is partnering with US-based Regent Craft to manufacture electric "seagliders" in the UAE, aiming to revolutionise coastal travel between Abu Dhabi and Dubai.
Regent Craft (Not an original concept- I covered Aotearoa-based Ocean Flyer’s original deal with Regent way back in Memia 2022.17:
“Startup Ocean Flyer announced its aim to purchase and operate a fleet of 25 electric ground-effect seagliders - promising to fly people on routes such as Wellington to Lyttelton in 1 hour for $60 a seat, starting in 2025.“
(Er, it’s 2025…!?)
🧮Future computing substrates
Finally, a whip through recent advances in computation platforms:
🧠💻Wetware-as-a-service Previously covered a few times since Memia 2023.42, Australian startup Cortical Labs has officially launched the world's first commercial "Synthetic Biological Intelligence" (SBI) system, the CL1, which fuses human brain cells with silicon hardware to create dynamic neural networks. The company offers energy-efficient "Wetware-as-a-Service" in the “Cortical Cloud” accessible to researchers globally. Early days, but the energy efficiency of biological neurons surely points towards this as a superior alternative to gigawatt-hungry silicon GPUs for AI processing…
Brain in a box (Cortical Labs) 
Human brain cells on a chip with input/output electrodes Vortion Researchers from the Universitat Autònoma de Barcelona have developed a new magnetic state called a "vortion" (magneto-ionic vortex) that enables unprecedented control of magnetic properties at the nanoscale, mimicking brain synapses with voltage-controlled nanomagnets.
Ocelot After Microsoft’s big Majorana quantum computing chip reveal last week, Amazon had a more low-key unveiling of its first quantum computing chip, “Ocelot”. After four years of development, Ocelot features innovative "cat qubits" that promise significant improvements in error correction— still the major challenge in quantum computing...
AWS ELMMs Researchers at Tsinghua University have developed Engineered Living Memory Microspheroids (ELMMs), a novel DNA data storage system that encapsulates bacteria containing data-encoded plasmids in hydrogel microspheres. The bacteria-filled microspheres enable room-temperature DNA storage with 100x hard drive density, allowing precise data retrieval through fluorescent sorting at speeds up to 0.55 MB/s.
🧵Thread computer MIT researchers have developed a groundbreaking single-fiber computer that weighs less than 5g and can be integrated into textiles for wearable computing applications. The fiber computer combines analogue sensing, digital memory, LEDs, processing capabilities and bluetooth communications within a machine-washable elastic fiber capable of over 60% stretch:
Gupta, N., Cheung, H., Payra, S. et al. A single-fibre computer enables textile networks and distributed inference. Nature (2025). https://doi.org/10.1038/s41586-024-08568-6
⏳ Zeitgeist
Once around the world trying to keep an eye on the ball rather than the player…
‼️Natural CO2 sequestration is down?
A new paper Natural sequestration of carbon dioxide is in decline: climate change will accelerate from Scottish climate scientists points to some pretty concerning implications if the modelling turns out to be correct:
“The rate of natural sequestration of CO2 from the atmosphere by the terrestrial biosphere peaked in 2008. Atmospheric concentrations will rise more rapidly than previously, in proportion to annual CO2 emissions, as natural sequestration is now declining by 0.25% per year.”
🌕🚀👻Blue Ghost on the Moon
Firefly Aerospace's Blue Ghost lander successfully touched down on the Moon on Monday at 3:34 a.m EST, the first successful demonstration of commercial lunar delivery capabilities for future exploration. The craft delivered NASA science and technology instruments for a 14-day scientific mission. Here’s the first image it sent back from the surface:
4 days earlier, the Firefly craft sent back this spectacular footage from lunar orbit:
🌊💩Flooded zone…
And as for the rest, another week in Trump:
F**k around: On Tuesday, the US Trump administration imposed 25 per cent tariffs on Canada and Mexico, and an additional 10 per cent levy on China. All three countries indicated that they intend to take retaliatory measures in response.
…Find out: As I write this, the US stock market is on a downward trajectory, at one point wiping out all gains since Trump was elected:
Trump Gaza Trump shared a bizarre AI-generated video envisioning his “vision” of a “Riviera of the Middle East” featuring Trump, Netanyahu and Musk. Jeezus.
DOGE escalates the Trump administration has directed US government agencies to prepare for additional layoffs as part of the Department of Government Efficiency (DOGE) initiative. Legal challenges to firings have created widespread uncertainty for government workers - meanwhile mass federal layoffs could reduce consumer spending.
Russia Rehabilitated? As I surmised last week, US Secretary of State Marco Rubio openly confirming that the Trump administration are aiming to “peel Russia off of China”. Hence:
The unprecedented Oval Office ambush of Ukrainian President Zelensky in front of the press - and the US has now cut off military aid for Ukraine. Europe is scrambling. (Looking forward: why would any country rely on US weapons systems from now on? AUKUS appears dead in the water, for one.)
US Cyber Command ordered to cease operations against Russia despite historical threat assessment.
(It’s almost as if the US is playing Checkers while China plays Go…)
Crypto reserve? On Sunday, Trump said that five cryptocurrencies would be included on the US Federal Reserve’s balance sheet, creating an American “crypto reserve” including bitcoin, ethereum, solana, cardano and XRP.
Edward Luce in the FT: Trump’s heist in broad daylight
Bitcoin fell 27% drop from its peak to below US$80,000 for the first time since November 11 last week, but boosted back up to US$87K after the crypto reserve plan was announced.
First US measles death in a decade highlighted risks in under-vaccinated communities. Maintaining 95% vaccination rates is essential for maintaining “herd immunity”… but you get the sense this ain’t gonna happen.
US Flu cases surge to 15-year high as vaccination rates decline
February 2025 hospitalisations exceeded any week since before 2009 with over 4,000 deaths since January 1. Again, vaccination rates have declined significantly since COVID-19, especially among high-risk groups.
Iran expands near-weapons grade uranium stockpile by 50%.
🎭Meme stream
Some light relief to end the newsletter with…
🎼AI does Mozart
You can now make classical music scores using open-source generative AI (Video via @dreamingtulpa. The library used is notagen if you want to try it out yourself.)
🤖💭Neo Gamma, plotting
That Neo Gamma household chores robot video… but with an altered soundtrack and darker colour palette. (@AISafetyMemes on form as always…)
🛑Swasticar
Appearing on bus stop billboards around the UK this week:
(Musk’s recent escapades are driving a sharp fall in Tesla sales in Europe).
😵💫Arguments for DOGE
However, you can see the arguments for, sometimes…
Crazy week, huh!
🙏🙏🙏 Thanks as always to everyone who takes the time to get in touch with links and feedback, it’s always appreciated.
Namaste
Ben