AI investment… go figure💸 reverse Turing🔄 biocomputing🧠💻 tech vs. state🖥️⚖️🏛️ 1X NEO beta👤 drones for wildfires🔥 paperclip maximizer reimagined🖇️ #2024.35
That's def a dude in a suit
Welcome to this week's Memia scan across emerging tech and the exponentially accelerating future. Thanks for reading 🙏
🏊This week I’m on holiday! Enjoying a week swimming around the coastline of the Greek Island of Symi… pretty idyllic (and hot!) As such, this week’s missive is a bit shorter and more impressionistic than usual…hopefully still provides your weekly fix!😇
ℹ️PSA: Memia sends *very long emails* (even when on holiday), best viewed online or in the Substack app.
🗞️Weekly roundup
The most clicked link in last week’s newsletter was: Fathom AI transcription.
ICYMI
The first print run of my book is nearly sold out… still time to swing by the ⏩Fast Forward Aotearoa store to purchase a physical copy (and limited edition merch while stocks last):
(Use code MEMIASUB10OFF for NZ$10 off your paperback order).
🏔️Nomadfest Suisse - next week
Also a reminder that next week I’ll be attending Swiss Nomadfest in the mountain town of Liddes in Switzerland from 8th-15th September. I will be giving a presentation on AI for Digital Nomads and taking part in the week’s activities with nearly 100 other nomads who are on the road right now! (If you’re in Europe in September there are a few spaces left - you can book using code BEN10 for 10% off).
🤖Robonews
Callaghan Innovation CEO Stefan Korn has been experimenting with AI1, building a daily news summariser AI app which includes an autogenerated summary image:
The text-to-image models still a little way to go, clearly… (the new open-weights Flux.1 image model is pretty good at rendering words correctly but both DALL-E (ChatGPT) and Midjourney still struggle.) They will get there soon.
Currently in development at Memia is a daily AI and tech news summary email for subscribers… alpha launch some time in the next month hopefully… if you’re interested in being an alpha tester and receiving this please let me know. Aiming to work with the amazing Sam Ragnarsson to develop a Memia AI image summary “house style”…
(Long-time readers may recall the concept of a Vibe Cube covered in Memia 2023.13 which is a variation on the same theme…)
📈The week in AI
AI news I’ve been filtering...
💸AI investment… go figure
The rationale behind continued massive global AI investment seemingly defies conventional financial analysis of Return On Invested Capital (ROIC):
Nvidia keeps hitting its straps Nvidia soared past analysts’ expectations *yet again* hitting $30.04 billion in revenue for Q2 FY25, marking a 122% increase compared to the same period last year. The GPU bonanza continues… but the stock fell back since as investors reconsider how likely it is that the firm can continue to keep up this run rate:
NVDA is basically *THE* bellwether stock for AI… - so the key question for us all to ask at this point: where is Nvidia’s future revenue growth going to come from? The stock is currently trading on a P/E ratio of over 52 🤯 — Is there really that much value to be had?
Sequoia’s David Cahn asks out loud:
(“Sovereign AI” is an interesting one…really?)
Ultimately for Nvidia to his these Wall Street growth targets, it requires more massive rounds of capex from the other members of the Magnificent 7 US tech giants… but what is the hope for any ROIC… ever… at this price?
OpenAI
bailoutnew investment round continuing the theme, Microsoft, Apple and Nvidia are said to be in talks to participate in OpenAI’s latest investment round, reputedly valuing the company at around US$100Bn. (This is the same company that is rumoured to be on track to lose US$5Bn this year). According to the FT, as part of the deal, it’s likely changes to the company’s novel non-profit governance structure would likely be forced through.Go figure… overall, total AI investment is difficult to quantify, let alone predict. Indeed there are numerous analyst reports which offer overlapping definitions of what constitutes “AI Investment”. Just a few of the main culprits:
Netbase Quid AI Index Report 2023 is used by Our World In Data to show a glut of AI investment in 2021
The HAI Stanford AI Index Report from earlier this year also covers last years boom time for GenAI companies raising investment:
1 year ago Goldman Sachs was projecting a continued upswing, with up to US$200Bn projected in 2025 (note the lack of comparing apples with apples… the chart below only shows private investment):
(That Goldman Sachs projection feels a bit out of date 1 year on, but so far I haven’t found any publicly available updated analyses of projected total AI investment this year.)
Add to this concerns that this is a fight-to-the-death: a "winner takes all" dynamic has become clearly apparent in the AI market over the last year, where just a few leading companies can be expected to dominate and capture the majority of customers and revenue. In this scenario, even AI startups with significant funding will struggle to compete against established tech giants with a massive moat. (Unless…. open source…)
*EITHER* we are in a bubble… Meredith Whittaker, President of Signal, the not-for-profit secure messaging app channels the vibe in an interview with Wired celebrating Signal’s 10th anniversary:
"I think this generative AI moment is definitely a bubble…You cannot spend a billion dollars per training run when you need to do multiple training runs and then launch a fucking email-writing engine. Something is wrong there."
…*OR*, one or more of the following hypotheses will eventually turn out to be correct (thanks Claude for summarisation assistance):
Genuine technological breakthrough leading to AGI: Recent AI advancements, particularly in areas like large language models, represent a true path to AGI with far-reaching implications for humanity, way beyond mundane concerns like ROIC (Return on Invested Capital)
Vast market potential: The addressable market for AI technologies spans virtually every industry, justifying large investments.
Strategic necessity: Investment in AI is crucial for companies to remain competitive, driving genuine demand rather than speculative interest.
Long-term value creation: Current investments may be justified by the long-term transformative potential of AI, even if short-term returns are not immediately evident.
(Another non-commercial dynamic is the escalating US vs. China military AI arms race …)
As usual I maintain a superposition on both sides of the fence…. but the investment dynamics definitely show signs of the 1999 dotcom bubble all over again…
🍓🗓️Strawberry… this year?
After all the (unfulfilled) prophesies of OpenAI soft-launching their reasoning AI code-name 🍓, The Information reports that OpenAI's Project Strawberry is now expected to become ChatGPT-5, is set to launch in Northern Hemisphere autumn this year with significant improvements in mathematical and programming capabilities.
(Non-paywalled writeup from Techradar here)
Perhaps related to the timing:
👁️AI product zeitgeist
The push to productize AI intensifies…
🔄Reverse Turing (1)
Soon AI captchas will require AIs to prove they aren’t human:
🖇️Paperclip maximizer - reimagined
Amazing what one person working alone with AI can now achieve… this is the SOTA as of August. (Watch to the end for a twist in the tale…)
(Named after philosopher Nick Bostrom’s classic Paperclip (now Squiggle) Maximizer AI thought experiment).
🔮[Weak] signals
Non-AI tech signals from near and far futures...
🖥️⚖️🏛️Tech vs state
Escalating tensions between censorious nation states and pan-national tech companies offering “free speech”:
Telegram vs. France Telegram founder Pavel Durov was released from arrest but placed under judicial supervision, having to put up €5 million bail, barred from leaving France and indicted on preliminary charges relating to Telegram's alleged noncompliance with inquiries into child abuse material and drug trafficking.
Edward Snowden was among many who robustly objected to the arrest at the time:
“The arrest of [Pavel Durov] is an assault on the basic human rights of speech and association. I am surprised and deeply saddened that Macron has descended to the level of taking hostages as a means for gaining access to private communications. It lowers not only France, but the world.“
X vs. Brazil A feud between Brazilian Supreme Court Justice Alexandre de Moraes and X owner Elon Musk has now erupted into a full-scale ban for X in Brazil. following a ruling from the Brazilian Supreme Court last week. The decision stems from X (formerly Twitter) 's failure to comply with local misinformation censorship regulations, particularly a court order to appoint a legal representative in Brazil after the platform closed its local office in mid-August.
The ban was implemented from Saturday (August 31) and affects over 22 million X users in Brazil. (Bluesky gained 500,000 new users shortly after). In addition, the court ruling imposes fines on individuals or businesses still accessing X through VPNs. Apple and Google have been given a five-day deadline to remove X from app stores and block its use on mobile devices.If the ban holds up, particularly the app store restrictions, then this will shift all X traffic away from apps and exclusively onto web browsers and VPNs (for those who want to risk it). Only a minority of users will go there… so X becomes fringe in Brazil. (Shows how significant Google and Apple have become as regulatory leverage points…)
Musk is in defiant mode, tweeting lots of strong opinions about the case. Plus this re: his other business, Starlink, which is launching mobile satellite internet soon and may enter Brazil whether the courts like it or not…
For reference, here’s the latest Starlink deployment:
Signal vs. Venezuela and Russia last month Signal announced that the app appeared to be blocked in Venezuela and Russia. Signal recommends users switch on its “Censorship circumvention“ feature…which still needs other big tech companies to fix their standards…
There have also been recent outages of YouTube in Russia, suspected to be state censorship.
Google “safeguards” for a different approach, Google is one of the few tech companies who censor themselves! As the US presidential election looms, Google is trying to play the good citizen by introducing additional “safeguards” on its generative AI products to try to prevent (…liability for…) misinformation:
Restricting election-related responses in AI products like Gemini
Providing “authoritative information” on voter registration, election candidates and political parties on Google Search results
Google Play will introduce a badge for official government agency apps.
Laurie Richardson, Google’s vice president of trust and safety, inadvertently pointed out the core issue in the company announcement… (eg. a private company acting as a public utility):
“Particularly for federal and state-wide elections, our users depend on us to provide reliable and up-to-date information on topics like current candidates, voting processes, and election results — and this new technology can make mistakes as it learns or as news breaks”
👤1X NEO beta
The most lifelike body form humanoid robot yet: NEO from Norwegian startup 1X came out of stealth. This demo is uncanny… is it real?
Here’s the backstory - “humanoid robots for the home”. It appears that yes, it is real… although the robot in the video *could* be operated remotely using VR rather than autonomous.
Some people aren’t convinced though:
Reverse Turing (2):
How long now until…
And a quick scan across the other links of the week:
🔥Drones for wildfires
A project bringing together Lancashire Fire and Rescue together with British self-flying cargo aircraft firm Windracers and AI and robotics scientists from the University of Bristol and the University of Sheffield has developed AI swarm technology and self-coordinating drones to detect and investigate wildfires:
⚡Energy
Japan to launch world's first steady-state nuclear fusion reactor? Japanese startup Helical Fusion has unveiled plans to launch the world's first steady-state fusion reactor by 2034, with an initial power generation capacity of 50-100 megawatts and an estimated construction cost of US$5 billion.
50-Year Battery?: BetaVolt has developed a coin-sized nuclear battery that can reportedly last 50 years. (But its low power output makes it impractical for consumer electronics like smartphones, which would require a battery weighing hundreds of pounds to function).
🧬Biotech
Scientists have discovered a groundbreaking protein called DNA damage response protein C (DdrC) in the bacterium Deinococcus radiodurans, which shows remarkable potential for DNA repair and cancer prevention:
🧠💻Biocomputing
FinalSpark, a Swiss tech startup, claims to have introduced the world's first living computer using brain cells for computation, marking a significant advancement in the field of biocomputing. The groundbreaking system utilises living neurons instead of traditional silicon chips and aims to massively reduce the energy consumption required for AI:
🧠Mind expanding
🎧Listening
Lots of time to listen to podcasts on my travels this week…here are a couple of the best:
Out of Africa… many times Dwarkesh Patel interviews the American geneticist David Reich, who specializes in ancient human genetics, on the latest DNA evidence on the origins of humanity. Absolutely fascinating conversation:
Former Taiwanese Digital Minister Audrey Tang talks to The Economist about How technology can strengthen democracy
Protein folding
Hypnotic animation of a reverse transcriptase that defends bacteria from viruses. (Sound on)
(Video via @maxewilkinson)
⏳Zeitgeist
Once around the world…not paying much attention to mainstream news this week…
🌡️14 degrees warming?
Not to put a downer on all this tech-optimism…. but new concerning new research reveals that CO2 has more atmospheric warming impact than previously thought - suggesting that doubling atmospheric CO2 levels could raise Earth's average temperature by 7 to 14 degrees Celsius, significantly higher than previous estimates:
🚗⚡Converting old cars to EVs
Rest of World reports on the increasingly popular practice in Latin America of turning petrol-fueled cars into electric ones, which can be four times cheaper than buying a new EV:
🧘Memetic savasana
Just one funny to share while I’m on a break…
🌈My AI hallucination holiday
Magnific founder Javi Lopez fed his recent holiday snaps into an AI video generator… with more-bizarre-than-usual results:
🙏🙏🙏 Thanks as always to everyone who takes the time to get in touch with links and feedback.
See you next week from the mountains of Switzerland!
Namaste
Ben
Would that every public sector CEO was so inclined…
That Witlowski paper on climate sensitivity is considered "unpersuasive" by climate scientists I know. Current range is quite worrying enough!