Fly connectome🪰 e/ac❄️ I-XRAY🕶️ memecoin supercycle🎭 wen neobanks?💳 telerobotic surgery🏥 wiretap on a wiretap📞 facepoke🥴 lipdub🫦 X-parrot🦜 #2024.40
Fusion ftw ☀️
Welcome to another weekly Memia scan across emerging tech and the exponentially accelerating future - 40th of the year! As always, thank you for being here.🙏
(Sorry for slightly delayed publication today…I’m finally on my way back to Aotearoa after nearly 9 weeks away and had to finish this off in Ho Chi Minh City departure lounge! …Looking forward to getting home with fresh eyes after experiencing first-hand what’s happening in different parts of the world.)
ℹ️PSA: Memia sends *very long emails*, best viewed online or in the Substack app.
You can also listen to this newsletter as a podcast discussion:
🗞️Weekly roundup
The most clicked link in last week’s newsletter (other than the NotebookLM-generated podcast) was the NotebookLM podcast hosts discovering they’re AI, not human—and spiralling Into existential meltdown.😅
Also a few follow ups from last week’s missive:
🌀Hurricane Helene aftermath
🎩 Gareth Renowden for pointing to this article from Yale Climate Connections: Four ways climate change likely made Hurricane Helene worse.
The fallout from Helene exposed the global semiconductor industry’s dependency on the small US town of Spruce Pine - which produces as much as 90% of the world’s scarce ultra-pure quartz, used to make the crucibles for producing high-grade silicon.
1 week after Helene…Hurricane Milton went from being a Category 1 to Category 5 in less than 12 hours and is hitting land as I type, projected to be the fourth strongest Atlantic hurricane in US history. Stay safe, folks in Florida.
❄️e/ac
Another week in the heat and humidity of South East Asia has made me acutely aware of the importance of air conditioning… a new report from UNEP explores the US$8 trillion opportunity for developing economies to avoid high costs of cooling from new, sustainable solutions:
🤝Work with me in 2025
<Very occasional self-promotion>
Five years of putting together the Memia newsletter each week has given me a unique perspective on how AI-driven change is accelerating. In particular, how AI is becoming increasingly relevant for those running real-world businesses outside the Silicon Valley venture-capital-hype bubble.
Over the last year I’ve worked as an advisor with leadership teams and boards across many diverse sectors including: energy, telecoms, infrastructure, logistics, financial services, health, law, agriculture, education, NGOs and government — as well as many technology firms.
Looking ahead to 2025…some ways you can work with me:
Get the most up-to-date AI Keynote for your conference or in-house event
Get a comprehensive AI landscape briefing for your Board or Exec team relevant to your industry, product or service
Bring in an external pair of eyes to help with shaping your organisation’s AI strategy
Get hands-on with AI capability building sprints for your team
Typical engagements range from one-off briefings, through 1-2 month AI strategy engagements to long-term capability building partnerships. If that sounds of interest to you, please reach out. Thank you!
</self-promotion>
📈The week in AI
The week's AI news and updates… trying to keep things short and sharp.
💸Non-profit for-profit
The AI bubble continues to inflate… are we there yet?
OpenAI confirmed that their latest funding round closed, raising US$6.6 billion at a staggering US$157 post-money billion valuation. (A significant increase from its previous valuation of approximately US$86 billion earlier this year, nearly doubling its worth).
The round was led by Thrive Capital, which invested around US$1.3 billion(!) Microsoft contributed approximately US$750 million, bringing its total investment in OpenAI to about US$13 billion. Other notable investors included Nvidia, SoftBank, and several VC firms including Tiger Global Management and Khosla Ventures. However, Apple decided to pass at the last minute…
Strings attached:
Investors can pull their cash if OpenAI does not transition fully into a for-profit entity within two years. That could get, er, dramatic.
Exclusivity clauses: according to inside sources, investors in this round are restricted from investing in certain competing companies, in particular: Anthropic, xAI, SSI (Ilya's company), Perplexity … and (one not a lot of people have heard of yet) enterprise AI firm Glean. (Non-compete clauses for investors still legal, apparently… and good luck stopping Nvidia spreading the love around…!)
Anyway, this round makes OpenAI the largest venture-backed company by valuation from Silicon Valley (surpassed by TikTok owner China’s ByteDance at US$220Bn, with SpaceX is also in the same league.)
Apparently OpenAI gives employees and investors PPUs - Profit Participation Units, not equity. But when will this firm *ever* make a profit? (Reminder: OpenAI is projected to make a loss of approximately US$5 billion this year, on anticipated revenues of around US$3.7 billion. Go figure.)
Emad Mostaque nails the economics of the OpenAI investment in one pithy tweet:
(Combine this with the firm’s headlong embrace with US National Security interests, is OpenAI the new Boeing? Also this week the company published a new “Raising Concerns Policy” — whistleblowing to you and me — which lays it out in broad daylight:)
“…OpenAI continues to distinguish between raising concerns and revealing company trade secrets. Given our technology's implications for U.S. national security, the latter (subject to the right to make protected disclosures) remains prohibited under confidentiality agreements for current and former employees.”
🧪AI and science
Two stories which talk about how AI is shifting science forward rapidly. (Pretty high level - who are the people to follow on the inside track of these advancements…?)
FT: DeepMind and BioNTech are creating “AI lab assistants” to help researchers plan experiments and predict outcomes. Google DeepMind is deep in creating a specialised science model while BioNTech introduced “Laila”, an AI agent built on Llama 3.1 which automates routine scientific tasks.
Nature: scientific researchers who helped to test OpenAI’s latest LLM, o1, say it represents a big step up for advancing science. In particular the new “chain of thought” has delivered significant improvements in scientific reasoning, outperforming PhD scholars on a challenging science test and impressing researchers with its enhanced capabilities in fields like quantum physics and coding.
🆕New research and releases
This has been a week of shipping… many significant new announcements to cover:
NVLM 1.0 Nvidia (newly invested in OpenAI) put the cat amongst the pigeons, releasing their own open source multimodal LLM series NVLM 1.0. This release includes the flagship model, NVLM-D-72B, which contains “only” 72 billion parameters but benchmarks on a par with Llama 405B:
So what’s Nvidia’s game here…? On the face of it, this is competing directly with their customers further up the stack… but maybe it just sets the “default model which comes with your H100” in the future. Every piece of Nvidia hardware you buy in future comes “pre-loaded” with NVLM and its successors…
OpenAI DevDay In addition to landing a shedload of cash, OpenAI also put on their second-ever Developer Day event. Minus Greg, Ilya, Mira and with Sam only coming on at the end for a fireside chat… nonetheless there was a lot to get through. No livestream (in-person attendees were asked NOT to record the keynotes) - however you can go to Youtube for a comprehensive summary from the Latent Space podcast.
Main announcements:
Realtime API (Public Beta): allows developers to create low-latency, speech-to-speech interactions using six preset voices. Basically integrate the voice functionality of ChatGPT Advanced Mode into your own apps… here’s a demo of it in use with an app called Firecrawl which uses the API to enable “chat with any website”:
Vision Fine-Tuning: Developers can now fine-tune the GPT-4o model using both text and images, making it applicable in fields such as medical imaging and autonomous vehicles
Prompt Caching: a cost-saving feature (following Anthropic who introduced it earlier this year) allows developers to reuse frequently used prompts, potentially reducing input costs by up to 50%.
Model distillation enables developers to fine-tune smaller models based on larger ones like GPT-4o
Meanwhile, let the implications of the pricing for Realtime API sink in: your very own intelligent voice agent for US$18/hr! (For comparison, US minimum wage is US$7.25/hr…)
All a very different vibe from last year’s DevDay when CEO Sam Altman was on stage with Microsoft’s Satya Nadella… here trying to make light of the last year’s ups and downs:
Canvas And then no sooner was DevDay over with, another announcement from OpenAI: ChatGPT Canvas, a new AI-first text and code editor with a separate window interface for more efficient project collaboration. Pretty much Claude Artefacts but with editing… or Google Docs with AI. Going up against the amazing AI-enhanced IDE Cursor most directly…
(Video: @OpenAIDevs)
Llama Stack Meta released their first full official distribution of the full Llama AI stack, packaging multiple API Providers into a single endpoint for developers. (Echoes of Meta’s very successful PyTorch strategy…)
And then we get on to a crazy week in AI video… wow!
Movie Gen Meta introduced Movie Gen, their most advanced genAI text-to-video and -audio tool to date, with powerful video editing tools built in. Here’s the showreel… people comparing it with OpenAI’s Sora in capability.
A few more example demos of usage:
(Although the research is announced, Movie gen is not yet available for public use due to ongoing refinements and high operational costs).
Pika 1.5 A new fun model from Pika Labs to play with - in particular “Pikaffects” - built-in special effects that “break the laws of physics”. I’ve seen a lot of people posting rolling up in plasticine “squish it” videos this week…
Excellent showreel:
Cake-ify: Hot air balloon
(Video: @ytjessie_)
Squish-it:Meme destroyer
(Video: @LinusEkenstam)
🥴Facepoke this has been blowing up on my wires this week. Facepoke is easy GUI-led animation and editing from a photo, created by Hugging Face engineer Julian Bilcke. Example below - just upload a photo to the space and away you go!
🫦Lipdub Likewise this, one of a number of AI tools which now let you get characters in a video to lipsync to a script:
(Clever video demo from @Uncanny_Harry)
And Heygen’s new Avatar 3.0 technology (covered a couple of weeks ago) are pretty mindblowing out of the box, now with “Unlimited Looks”. (Coming soon: me narrating the newsletter…)
(Video via @Linus Belunias)
Meanwhile more signs of the rapid convergence of Hollywood and AI:
Runway Studios is hiring:
Text-to-show Jars.ai has a very early “text-to-show” generator… this is pretty basic, but if you think what NotebookLM can now do for audio podcasts… not long until entire TV shows are generatable off a prompt? Will they be noticeably worse / the same / better than TV shows with human scriptwriters?
Submerged FINALLY …. Apple released a trailer for the first scripted film in Apple Immersive Video - released tomorrow. A new medium emerges…
🎙️Listening about AI
Finally…. a few podcasts and lectures with AI luminaries which I’ve been listening to this last week:
Dwarkesh Patel interviews Dylan Patel (SemiAnalysis) and Jon (anon - Asianometry) discussing the modern semiconductor industry… some deep insights inside here, energetic and fun to listen to.
Lex Fridman interviews the team from Cursor - I’ve been using this tool for a few weeks now and its mindblowingly good at writing and fixing code for me.
Erik Brynjolffson, director of the Stanford Digital Economy Lab gives an accessible lecture on how AI is impacting businesses. I first saw Erik speak in 2019 at the COGx Festival in London and his insights then were years ahead of others, if a little vanilla.
🔐Crypto sidequest
I haven’t covered happenings in crypto for quite a while… sort of lost touch (and interest, tbh). Anyway, a few things passing across my feed this week worth sharing…
🎭Memecoin supercycle?
Thanks to my friend Oli who enthusiastically sent me a link to a recent talk by “Memecoin” evangelist Murad Mahmudov at the recent Token 2049 conference in Singapore. (Perplexity tells me that…)
“Memecoins are a category of cryptocurrency that primarily derive their value from humour, internet memes, and cultural phenomena rather than from serious technological innovation or utility.“
(Note the word “serious”).
Murad’s breathless assertion is that a new wave of Memecoins are poised for substantial growth relative to the rest of crypto, potentially reaching a market cap of US$1Tn from today’s (very generous estimate of) US$49Bn. And this growth is from some opaque *value* derived through community, identity, and entertainment, rather than any conventional model of technological utility:
I watched all of this and I just can’t help thinking we’ve been *exactly* here before, only a few years ago with the 2017 ICO (“shitcoin”) wave. I was there. And then… NFTs. And then… FTX. So far retail crypto is just… grift.
But Murad clearly believes in his theory… and pushes back against blinkered utilitarians like me:
Whole different world, eh? The argument is that the next generation of investors, effectively shut out of the “real” economy unable to get a well-paying job or buy a home, will instead turn to putting their money into memecoin tokens instead of Tradfi savings and investment products? I am far, far from convinced.
(The darkest cynic in me says that this emerging Memecoin space is just another bunch of grifters planning to get in early to own a piece of the next casino… )
But at the same time, there is *something* in here… some of the memes and in-jokes are marginally funny and, even, satirical. Throughout history, people have paid for entertainment and information, there will always be information markets… perhaps in future “owning” a BONK NFT is as much of a monkey-status symbol as a Rolex or a Ferrari was…
So at this stage:
No “serious” utility function that I can see. Murad’s definition of the motivations for memecoins is too opaque for me.
(However, given that the entire global economy is a machine for turning billions of barrels of oil into microlitres of dopamine… the utility function of GDP growth isn’t “serious” either…!)
I think crypto token market mechanisms are far too primitive for facilitating real-world memetic value exchange...
Has all the dynamics of another imminent wave of pumping and dumping
Emad was also at the same conference. Has a not-dissimilar take:
💳Wen Neobanks?
Given all of the preceding… the primary use case for crypto *since forever* has been peer-to-peer payments and self-custody of financial assets. But it still isn’t hitting the mainstream and integrating seamlessly with fiat currencies that are still used day-to-day…
I’ve seen this claim so often:
But…maybe we are nearly there…. a couple of standouts from the crowd:
“The world's first self-custodial Visa Debit Card, linked to a Safe Smart Account. Spend your crypto like you spend cash anywhere worldwide and get up to 5% cashback.”
Nice looking plastic:
Zeal Wallet: founded by an ex-Revolut team, a fully-functional, self-custodial crypto wallet with seamless on/off ramps for both crypto and fiat payments and other products:
Recent interview with Zeal co-founder Hannes Graah, pretty illuminating on what’s coming:
🕵️♂️Odds-on Satoshi
Finally, HBO’s new documentary Money Electric: The Bitcoin Mystery, due to go out yesterday US time, has reignited significant interest and speculation about the identity of Bitcoin's creator, Satoshi Nakamoto. Money Electric’s director, Cullen Hoback, claims in a post on X to have worked it out:
Speculation initially led to Len Sassaman - who died only a few months after the original Bitcoin whitepaper was released. (A heartfelt tribute article from 2021 told his story… covered in Memia 2021.12).
However, at the time of going to press, cypherpunk Nick Szabo is ahead on Polymarket… I guess we’ll have to see if the mystery is actually solved…!
🔮[Weak] signals
Non-AI tech signals from near and far futures...
📞Wiretap on a wiretap
Oops. If you open a backdoor to your network… inevitably something like this will happen…
Chinese hackers linked to the Salt Typhoon group breached networks of major US broadband providers including AT&T and Verizon, potentially accessing sensitive backdoor wiretap systems being used by US intelligence agencies to spy on, er, US citizens. Details are scant right now but the WSJ sums up:
“For months or longer, the hackers might have held access to network infrastructure used to cooperate with lawful U.S. requests for communications data…The hackers appear to have engaged in a vast collection of internet traffic from internet service providers that count businesses large and small, and millions of Americans, as their customers.“
In Europe, Hungary is using its rotating presidency of the EU to push reworked “Chat Control” legislation again which aims to give governments the right to scan the contents of encrypted messaging apps… what could possibly go wrong…
🕶️I-XRAY
(We all sort of knew this was coming, in a Black Mirror kinda way…)
Builders AnhPhu Nguyen & Caine Ardayfio have folded together a bunch of open-source technologies, internet databases together with Meta’s smart glasses… creating a compelling demonstration of how any expectations of anonymity and privacy in public spaces are… artefacts of the past. Check this out:
(The creators have published a document with information on the tools used and details of how to request to remove yourself from various image / people search databases… good luck with that.)
🪰Fly connectome
A significant neuroscience breakthrough was published last week: a consortium of scientists known as FlyWire (previously covered Memia 2023.10 and 2023.37), co-led by neuroscientists Mala Murthy and Sebastian Seung at Princeton University, have created the most comprehensive map of a fruit fly's brain to date. The "connectome" includes nearly 140,000 neurons and over 54.5 million synapses which enabled the creation of a computer model of the entire fruit-fly brain, which accurately predicted fly behaviour. Although there are significant limitations — the map is only based on a single female fly and only shows chemical synapses — this is another major step towards the goal of reverse-engineering organic brains into software.
(Order of magnitude comparison: human brains contain around 86 to 100 billion neurons and a staggering 100 trillion synapses…)
🏥Telerobotic surgery
World's first intercontinental robotic surgery: In June, a Chinese surgeon successfully performed the world's first robotic surgery from Rome, removing a prostate lesion on a patient in Beijing, 5,000 miles away, using advanced telesurgical technology and a robust 5G network to overcome signal latency issues.
☀️Fusion ftw
Solar power now provides nearly 6% of global power generation. Which direction is this heading:
Looked at another way… (a nice thought-provoking shitpost from Michael Liebrich…) you could say that Nuclear Fusion is now >5% of global power generation:
(So…big picture… what is the advantage of spending $billions trying to build relatively tiny fusion reactors down here on Earth…?)
Terrestrial Solar investment continues to explode… just a few solar stories:
Raygen Australian startup RayGen has opened its groundbreaking solar and thermal storage project in Carwarp, Victoria, combining 4MW of high-efficiency photovoltaic solar with 2.8MW and 50MWh of long-duration storage.
The Australian Renewable Energy Agency (ARENA) has committed an additional AU$10 million to RayGen, bringing their total investment to AU$38.4 million since 2012. RayGen's innovative technology uses tracking mirrors to focus sunlight onto high-efficiency PV modules, converting one-third of sunlight into electricity and two-thirds into heat, which is then stored in two water pits for thermal energy storage.
An example of a solar-covered car park - now mandated in France:
Gravity-based solar cooling Saudi Arabian researchers have developed a gravity-powered cooling system for solar cells that harvests water from air without electricity, potentially doubling collection rates compared to existing solutions.
No more nuts and bolts?
3D-printed nickel-titanium alloys promise stronger structures without nuts and bolts: Researchers from Texas A&M University and Sandia National Laboratories have developed active interlocking metasurfaces (ILMs) using 3D-printed shape memory alloys, potentially revolutionizing mechanical joint design across industries with stronger, more adaptable structures.
🤯Negative time?
Quantum physicists at the University of Toronto say they have have observed photons exiting a material before entering it, suggesting evidence of "negative time". (The paper is still under peer review…)
⏳Zeitgeist
Once around the world outside tech…
🚯Time for action on microplastics
The surfeit of plastics in our environment is reaching the top of my Polycrisis radar…particularly on my recent trip around Asia where I became hyper-aware of how much plastic drinks packaging is used once and then just thrown away into the environment.
Some facts and statistics… from Unesco Ocean literacy website:
“Plastic waste makes up 80% of all marine pollution and around 8 to 10 million metric tons of plastic end up in the ocean each year.
Research states that, by 2050, plastic will likely outweigh all fish in the sea.
In the last ten years, we have produced more plastic products than in the previous century.
The EPA (Environmental Protection Agency) has stated that basically 100% of all plastics human beings have ever created are still in existence.
Plastic generally takes between 500-1000 years to degrade. Even then, it becomes microplastics, without fully degrading.
Currently, there are about 50-75 trillion pieces of plastic and microplastics in the ocean.
This plastic either breaks down into microplastic particles, or floats around and ends up forming garbage patches.“
This video from University of Plymouth showing microplastics mixed in with microscopic plankton sample shows just how much accumulation has happened:
(Source: @planktonpundit)
What plastic does to our bodies - a discussion on how plastic is increasingly linked to numerous serious health conditions from cancers to heart disease. And it’s changing our DNA—and now babies are being born “pre-polluted”.
A recent study showed that microplastics are accumulating in the human brain.
Here’s what’s being done about it: Why we need a strong global agreement on plastics pollution.
“Twenty years ago, scientist Richard Thompson sounded the alarm on microplastics pollution. Now, as understanding of the problem has grown, he says it is critical that international negotiators produce an effective plastics treaty when they meet next month in South Korea.”
🧠Mind expanding
Two pieces which have got my mind ticking over this week:
⚠️Anthropocene traps
Evolution of the polycrisis: Anthropocene traps that challenge global sustainability - a paper published by Peter Søgaard Jørgensen et al. at the end of last year, is a systematic analysis of the “evolutionary traps” humanity increasingly faces from new technology, global and structural constraints.
“The interconnected, global challenges of the Anthropocene lead to the question of whether we as humans could be on the verge of being, or already have become, locked into some form of undesirable trajectory with persistent crises and growing negative impacts on human well-being. Could the current Anthropocene trajectory be a trap that modern industrialized societies are naive to, not unlike seabirds feeding on deadly marine plastics, lacking the capacity to distinguish them from nutritious marine plankton?“
Most concerning: of the 14 “traps” identified, 12 are assessed to be in an advanced phase with high risk of lock-ins and negative impacts on human wellbeing.
(🎩 Matt Boyd for the pointer in his recent writeup Lessons from the Cambridge Conference on Catastrophic Risk 2024.)
✍️Creative writing… as a weapon
Can creative writing help America win wars? The Economist explores an unconventional approach to military strategy, using creative writing to prepare for future conflicts:
“Their jointly written novel, “Ghost Fleet”, was published in 2015. It’s set in the “not-too-distant future”: there are hologram maps, liquid body armour and “stim” tabs to keep tired soldiers awake, but the world it describes is recognisably our own.
The book opens with a sequence of shocking events: Chinese astronauts zapping American satellites with a space laser, Chinese cyber-troops smuggling malware onto the American army’s servers. With Russia’s help, China disables America’s communications and weapons systems, then invades and occupies Hawaii. It ends with a showdown in the South China Sea.
“Ghost Fleet” rang alarm bells in military mess halls and the corridors of power. A fighter pilot attending the workshop told me that he was urged to read it by a CIA agent. It has appeared on reading lists for the USAF and Britain’s Royal Air Force (RAF). The White House, the CIA and several other government agencies invited Singer and Cole to brief them; so did the Australian Parliament and the RAF.
It wasn’t long before people started asking the two men to give their white papers the “Ghost Fleet” treatment. The problem, they were told, was that people don’t want to read long reports written in turgid prose. As well as running workshops, Useful Fiction distils the research and lessons from say, a briefing on quantum tech or paper on cyber-security strategy, and works them up into fictional vignettes.”
Scenarios are definitely one of the most useful tools for foresight. (🎩 Roger Dennis for sharing).
🧘Memetic savasana
And time to wind down with some eclectica from around the interwebs…
🦜X-Parrot
Not the first time I’ve used this meme… but the best variation yet. (And yet *still* I’m there more than anywhere…)
🔄Amazing in motion
From 2014, TRANSFORM is a piece of “dynamically moving furniture” which responds to human input and can remember user preferences. A fascinating prototype… I wonder where this research has gone to since…?
👓Human connection
On-point response to Meta’s Orion announcement from last week from ITP’s in-house cartoonist Jim:
🖼️Déstructuré
Exploring new forms of art made with AI ,by Erwann Million:
🙏🙏🙏 Thanks as always to everyone who takes the time to get in touch with links and feedback.
Namaste
Ben