Sam Kenyon retweeted
Self-replicating, multi-agent AI systems are an overlooked existential risk.
I want to lay out why I believe this, via example.
Each dot is a simple creature with a randomly evolved NN brain.
There is no training; yet complex, unpredictable survival drives always emerge.
1/14
Sam Kenyon retweeted
Up until now, people have been smart enough for society. These days we are, as it were, running out of cognitive fuel. We’re not sufficiently cognitively complex to productively relate to the society that we ourselves have created.
Sam Kenyon retweeted
My dispatch from the largest citrus graveyard in America, the Orange necropolis, where, seemingly unbeknownst to most, the iconic breakfast beverage of the great American century—Orange Juice—finally ran dry. slate.com/business/2026/04/f…
The countdown begins.
Teams at @NASAKennedy have arrived to their stations at the Launch Control Center. We are about 48 hours from the launch of the Artemis II mission around the Moon. go.nasa.gov/4bHcwzx

ALT NASA\u2019s Artemis II Space Launch System (SLS) rocket and Orion spacecraft are seen atop a mobile launcher at Launch Complex 39B, Monday, March 30, 2026, at NASA\u2019s Kennedy Space Center in Florida. The orange core stage of the rocket stands between two white solid rocket boosters and is topped with the white Orion spacecraft. The rocket stands in front of the trussed structure of the mobile launcher. The sky is blue and cloudless in the background. Credit: NASA/Bill Ingalls
Sam Kenyon retweeted
Boop there she is
@NASA_Events
Sam Kenyon retweeted
Testing microgravity with monkeys. Chesley Bonestell's illustrations for Colliers magazine (1952)
Sam Kenyon retweeted
Replying to @SamVanivray @klara_sjo
We shouldn’t confuse current LLM limitations with theoretical limits. Kinetic intelligence requires constraint grounded learning, not just token prediction. That’s an architectural question, not a philosophical impossibility.
Sam Kenyon retweeted
CES, the world's largest tech show, just concluded in Las Vegas.
With robotics and AI front and center, it was supposed to show how far autonomy has come.
Instead, it exposed a much harder reality.
Here are the top highlights: (CES without the B.S.)
Sam Kenyon retweeted
An autonomous @rivr_tech delivery robot being tested in Pittsburgh. Snow, ice, hills and stairs everywhere: it was genuinely feeling dangerous as a human to go outside at times. Really a great stress test for the robots.
Sam Kenyon retweeted
I believe DARPA just launched the most audacious bioengineering program of the decade & nobody’s talking about it loud enough
I spent hours reading through the Generative Optogenetics program documentation & honestly I’m still processing the implications of what they’re trying to build
this is about creating molecular machines that can be expressed in living cells to translate optical signals directly into DNA and RNA sequences on demand
think about what this means…. you shine specific wavelengths of light at a cell & it writes its own genetic code in realtime without any external DNA delivery no viral vectors no CRISPR cassettes just light triggering nucleotide incorporation based on photon wavelength
this is a FOUNDATIONAL shift in how we interface with biology bc right now if you want to deliver genetic instructions to cells you need massive supply chains synthesis facilities cold storage delivery mechanisms that break down over distance
DARPA GO wants to turn living cells into programmable biological systems where you can transmit genetic information massively via optical signals & the cell synthesizes the exact sequence you need using its own machinery
the implications are absolutely wild :
single-cell spatial resolution temporal precision to deliver different messages sequentially remote scalable dissemination of genetic instructions
Imagine regenerative medicine where you shine light patterns on damaged tissue & cells reprogram themselves in situ
manufacturing where bacterial cultures synthesize complex molecules on command via optical programming
agriculture where crops adapt their genome in response to environmental light signals…
what fascinates me most is that this high risk program acknowledges we’re building a direct interface between computers that design genetic sequences and living cells that execute them
we’re not just editing genes anymore we’re creating a bidirectional communication protocol between silicon & carbon where light becomes the universal language
If GO succeeds it unlocks programmable biology at a SCALE
we’ve never seen before this is how you get to foundational capabilities for extended spaceflight for on demand therapeutics for manufacturing without brittle supply networks
this is the kind of moonshot that changes EVERYTHING
Sam Kenyon retweeted
The US Navy has managed a total of 273 nuclear reactors, 6200 reactor-years, over 177 million miles, averaging 4 new reactors per year over 70 years.
They have done this with a perfect safety record.
Zero accidents.
Zero injuries, zero deaths, zero environmental pollution.
Sam Kenyon retweeted
I misread a passage in the article.
It’s actually ductile at room temperature, melting point as high as 2000 degrees Celsius and it oxidises slowly even at super critical temperature ranges.
So, it’s actually way cooler than I earlier expected.
@atomalom thanks for correcting me
Sam Kenyon retweeted
The US is far and away investing more in AI and AI infrastructure than any other nation
The US currently has more total AI compute than all other countries combined
🇺🇸🤖
Sam Kenyon retweeted
Yasa (owned by Mercedes) recently smashed their own record for motor weight, size and horsepower. This production intent motor weighs 28lbs and produces 1,000hp!
Sam Kenyon retweeted
i thought it's mostly TPP, but it's actually mostly SiO2 + SU8 with EBL and photolitho, and a-Si as sacrificial layer for releasing the structure. The grating is a-Si then capped with SiO2.
Wang2025: [Microscopic geared metamachines](doi.org/10.1038/s41467-025-6…)




































