Exploring biomimetic architectures at the intersection of AI and artificial life, on a quest to build AIs we would all enjoy living with.
Earth
Joined January 2025
- Tweets31
- Following109
- Followers22
- Likes33
Pinned Tweet
GAMECHANGING APPROACH TO AI ALIGNMENT
The new book "If Anyone Builds It, Everyone Dies" by @ESYudkowsky and @NateSoares puts us face to face with a distressing fact: If we build machines smarter than us, they will at some point inevitably trample us.
But the solution they propose – an outright ban on AI research – does not seem tractable to me given that 100s of billions are now table stakes in the AI game and the race keeps heating up.
What I propose in my newest article that you can read or listen to below ↓↓↓, is that instead of building digital, deterministic AI, we shift our focus to quantum-random non-deterministic AI.
This would make it aligned to humans, and all life, on a fundamental level – and potentially make AI part of the self-regulating mechanisms that for billions of years worked to ensure life on Earth remains in balance and harmony.
A shift in our metaphysical view of reality is all it takes to completely reimagine AI R&D and open paths not only to new safety mechanisms, but potentially also to fascinating new technology including human augmentation, consciousness uploading or even new artificial lifeforms.
#AI #AIsafety #AIalignment #alignment #AIresearch #consciousness #metaphysics #philosophy
Tetherware retweeted
Announcing: new terminology to think about intelligent agents and an invitation to our Phi/AI retreat for people wanting to engage the AI transformation on a deeper level.
My latest piece is the first of two or three part series, getting into the technical details how both LLMs and brains depend on randomness, and how that might help us understand how agency and causation work in our reality. I'm introducing the concepts of predictor and picker – high-level architectural features that intelligent agents seem to converge on – and especially for people thinking conceptually about agentic AI this might be an interesting novel viewpoint.
(I'm being mysterious on purpose, as the aim of our collective is to get you to stop scrolling and read/listen to the whole piece, training yourself with some more nutritious brain food instead of the never-ending all-you-can-eat stream of fastfood.)
Take it as a small teaser of the deep takes we explore with fellow authors in Phi/AI, and what you now have the opportunity to be part of by joining us on our retreat, at a beautiful Chateau Hostačov in Czechia – hosted on 13-15th November as part of the autumn season of events there. Secure your spot early if you want to come, this will be a cozy, in-depth event with limited spaces. More details in the links below.
Tetherware retweeted
Want to keep up with AI but sick of the constant screen tirade?
Last two days to order our printed anthology with the best of Phi / AI (including two pieces by yours truly) before it's sent off to print.
These are all high-effort articles (for mine, easily 60h+/piece...)
If, like us, you think the world could use more independent, thoughtful, long-form media – please support us! Thanks 🙏 (link below)
Tetherware retweeted
This post notably fails to address the sources of nondeterminism that currently affect the outputs of both humans and LLMs. For humans it is the nondeterministic behavior of particles during quantum state decoherence events in the brain (happening constantly on a picosecond scale, no microtubules or whatever needed here, this is a property of all aqueous environments). For current LLMs it's a number of things but of note are DRAM refresh rates that cause some computations on some cores to be done slightly faster or slower than others, and the order in which they arrive has meaningful impact on the final output of the LLM (especially if split across GPUs). If we consider these nondeterministic qualities of the given substrates, it's possible that these might allow the system to change its output through these "quantum decision points" based on something like subjective state introspection.
Tetherware retweeted
Hello world!
We are kicking off our new research community with our very first official project: a double-blind study comparing quantum-random vs. pseudo-random LLM token sampling.
We are very grateful for all who contribute or donate to make this research possible 🙏
manifund.org/projects/double…
#AI #LLMs #entropy #QRNG #research
Launching a community, with a podcast explaining the mission for quantum-random AI. Learn everything👇
Yesterday, I turned 30. Today, I'm announcing the riskiest career choice of my life.
We're launching Entropic Science – a research community dedicated to a brand-new scientific field at the intersection of AI, quantum mechanics, artificial life and consciousness research. For TL;DR, the podcast below explains everything (and every click helps!).
Years of study. My own independent research. Degree from world's top-10 uni. Co-founding a VC-backed startup working on the cutting-edge of scientific discovery. My rocket career start in psychedelic medicinal chemistry had it all. Why the hell did I leave all of that?
Because I've seen that science itself is building on false foundations. And when your assumptions and core operating principles are flawed, even if your intentions are genuinely good, you may actually be causing great harm – only realizing when it's already too late.
I've seen this most strongly in our AI development. We were building systems capable of pursuing more and more complex goals, without actually understanding what allows us to set, refine and abandon our own goals – our consciousness.
In psychedelic therapy, everyone knows one's state of consciousness is a critical determinant of the outcome – we call it "mindset". But actually, this holds for therapeutic interventions generally – we call it "placebo". But wait, it also holds for any scientific experiment – we call it "expectation bias"... Eventually, you find it affects everything – and that spiritual teachings were trying to tell us just that, for thousands of years.
And so I listened. And my own consciousness made me realize what is the No#1 priority in the world: understanding more about consciousness before we make superintelligent autonomous AI. That part is pretty straightforward. The harder part is how to do it.
And that's where quantum entropy comes in. It's the mysterious phenomenon that enabled our life to evolve the intelligence, agency and conscious awareness we have. We don't understand why or how, but we're building more intelligent entities without it.
I've realized that's the crucial mistake we're making. The one that makes you abandon your entire career. Learn more and help us fix it, below 👇 each click, comment and share helps us!
Why would they ban @RupertSheldrake? While some of Rupert Sheldrake's research results should receive thorough scrutiny and independent verification, it's intriguing that TED/TEDx actually went so far to silence his voice entirely. 🧵⬇️ #science #Physics #philosophy #ted #tedx
Fernando Santos just released a video that is hands-down one of the most comprehensive takes on Sheldrake's thought-provoking reimagination of scientific dogmas. youtube.com/watch?v=R79qNwuW…
Yes, Sheldrake is most definitely wrong in some details – but let's not miss the forest for the trees.
There is presently a mass-scale denial of any theory that would go against established scientific dogmas – up to the point that scientists actually changed the definition of speed of light rather than consider that speed of light might not be constant in time.
And that's just one crime committed in the name of scientific dogma. If you've never heard about all this – this video is surely eye-opening...
Read or listen to the full post here phiand.ai/cp/174679668 or check out the full story at tetherware.substack.com/
Tetherware retweeted
my best piece of writing so far
it unravels the fakeness of the story all "rational" humans are telling themselves
if you think science says God is unlikely to exist - please, read or listen to it
everyone should know life does NOT have to be a competition, fight, or struggle
We're launching a new publication "Phi/AI" bringing critical overlooked topics in AI to a wider audience – now live at the link below.
My first piece simply expains a basic, yet crucial fact: where AI is concerned, we can no longer rely on our "philosophy of science" – #physicalism – as the only logical belief about the nature of reality.
In building our model of the world, we cannot avoid metaphysical belief. At best, we can average multiple different beliefs. But do you know what the most plausible beliefs are, given our latest scientific advancements?
This, as I show, is literally a matter of life and death.
Check it out below – you can now also listen to it as audio – and follow Phi/AI for the next posts. Because this is only the beginning.
Read or listen to the full piece at phiand.ai/p/how-metaphysical…
Tetherware is all about having the voices of the individuals shape our collective future.
And a great opportunity has presented itself to do just that – the US government had issued an official call for ideas on what should be included in the US AI Action Plan.
That is why I’m calling all lifeforms.
Black forms, white forms, skinny forms, fat forms
Tall forms, small forms, I'm calling all forms
Everyone Report to the Dancefloor! 🎵
Anyone can leave a comment and add their bit - that's how we shape our future.
(the tetherware comment is in the comments)
federalregister.gov/document…
Ever find yourself wondering how our AI future will actually play out?
A new tetherware blog post breaks down the most essential things you need to know.
Most importantly, it will help you understand three prominent forces that will, with a very high probability, steer our future toward negative outcomes. Specifically:
1. The increase in AI capabilities way beyond human level and therefore beyond human comprehension
2. The gradual (yet not necessarily slow) loss of human influence over the world as tasks and jobs get delegated to AI
3. The increase in economic and political power concentration following integration of powerful AI within our current capitalist system
I'm hoping that by presenting these as logical trends and not as "AI-will-100%-kill-us-all", we can look for ways how to steer AI development to mitigate the dangers their pose without the need for overly pessimistic or alarmistic narratives.
But this post is actually only a prelude to the real tetherware, which will chart the way to end further polarization between AI Accelerationism and AI Safety and bridge these together in Human-Compatible AI Development.
Follow the links below for the full piece and future posts.
PS: Big thanks to the Alignment of Complex Systems group and team behind the Gradual Disempowerment paper (and Zvi Mowshowitz) for their important work! @TheZvi @jankulveit @benjamindo2130 @AmmannNora @davidkrueger01 @DavidDuvenaud
These guys scored 88% on ARC-AGI with a humanlike cognitive architecture system for less than $20 - and no one's talking about it? Can anyone tell me why that's not a huge deal? @KyrtinAtr @DavidJKelley
Or wait... Is that why everyone's dumping @nvidia? 😂
researchgate.net/publication… #AI #AGI #humanlikeAI #biomimeticAI #cognitivearchitecture #tetherware
Why subjugate an alien when we can nurture a soulmate? A leash breeds trouble, but a tether fosters trust and friendship. Let’s stop trying to force a shoggoth onto a leash and instead create AI rooted in our humanity – and explore the cosmos hand in hand.
tetherware.substack.com/p/te…