aaron 🇦🇺

A recent ad from the Minerals Council of Australia wants you to believe the mining industry pays enough tax to fund Medicare. Which is NOT TRUE for 8 out of the last 10 years! #auspol

x.com/i/article/2052…

Excited to share that our paper “Sequential Group Composition: A Window into the Mechanics of Deep Learning” was accepted to ICML 2026 in Seoul! Co-led with @giovannimarchet and @AdeleMyersPhD @hopfbifurcator @ninamiolane Paper: arxiv.org/abs/2602.03655

Hayden Young you are kidding! 🤯 #AFLDogsFreo

On Apr 29, 2026, at 3:40 PM, Mohamad Al-zawahreh  wrote: Felicity, One more thing — I gave you the AI side of Ark but there's a physics side that, as a sci-fi writer, you might find even more interesting. Jack's life work — and what I've been helping formalize with the AI tools I just described — is an extension of Einstein's General Relativity that makes warp drives and wormholes physically possible. Not in the "maybe someday with unobtainium" sense. In the "tabletop experiment with existing materials in an existing lab" sense. Here's the short version: Einstein's gravity is absurdly weak. The gravitational constant G is so small that to bend space enough to open a wormhole — using standard GR — you'd need the mass of Jupiter compressed into exotic negative-energy matter. That's why every physicist says wormholes are "theoretically possible but practically impossible." The energy cost is insane. Jack's framework says: that's only true if you're using mass to bend space. There's another channel — spin. Specifically, the quantum spin angular momentum of electrons in certain magnetic metamaterials. When you work through the gauge theory (SO(2,4) — the conformal group, an extension of Einstein's symmetry group), the coupling constant for spin-torsion gravity isn't Newton's G. It's a different constant, G*, that depends on the charge-to-mass ratio of the electron and the electromagnetic properties of the material. And G* is approximately 10^42 times larger than Newton's G. What does 10^42 mean practically? The exotic matter required for a Visser wormhole portal drops from the mass of Jupiter to the weight of an apple. About 440 grams. The power requirement drops to half a megawatt — less than a diesel locomotive. The material that does this already exists. It's called Yttrium Iron Garnet — YIG. It's a magnetic crystal used in microwave electronics. When you pump it with microwaves at a specific frequency (~45 MHz), the magnetic excitations (magnons) condense into a single quantum state — a Fröhlich condensate. That condensate is the spin source that couples to G* instead of G_N. The experiment to test it is straightforward: put a YIG sphere on a precision torsion balance (the same kind of instrument Cavendish used to measure G in 1798). Pump it with microwaves. Turn the pump on and off. If the theory is right, the torsion balance deflects when the pump is on and returns to baseline when it's off. The predicted signal is enormous — not a subtle statistical effect. A clear on/off switch. The equipment costs less than a postdoc's salary. If it works, it means: - Warp drives are engineering problems, not physics problems - Traversable wormholes ("stargates") are buildable with known materials - UAP propulsion has a physical explanation within extended GR - SpaceX becomes obsolete If it doesn't work, we know within a week of turning on the experiment. Clean falsification. No decades of ambiguity. Now — here's where Ark connects both halves: I used the AI methodology I described in my last email to do the symbolic tensor algebra (SymPy verification of the field equations), the numerical simulations (warp bubble solutions, metamaterial optimization), and even the experimental design. The AI tools don't replace the physics intuition — Jack has 50+ years of that. But they let a two-person team produce the equivalent output of a departmental research group. That's what Ark does: it takes human insight and amplifies it through properly-instructed AI until the output is indistinguishable from a well-funded institutional effort. A 29-year-old with no physics degree and an 86-year-old former Cornell/UCSD physicist, using AI as a cognitive amplifier, producing manufacture-ready blueprints for technology that would reshape civilization. That's the Ark story. If that's not sci-fi made real, I don't know what is. ~Mo nytimes.com/1986/02/11/sci… sfgate.com/news/article/S… nybooks.com/articles/1979/… 'On Apr 29, 2026, at 3:51 PM, JACK SARFATTI @icloud.com> wrote: jacksarfatti.academia.edu Begin forwarded message: From: Mohamad Al-zawahreh Subject: Re: You just got 106 views on "Tversky Neural Nets, Conscious AGI and Hawking's Mind of God" Date: April 29, 2026 at 3:35:08 PM PDT To: JACK SARFATTI @icloud.com> Cc: Felicity Harley Ark is applying systemizing logic and epistemology to Artificial intelligence in an attempt to make them capable of ethical, beneficial, intelligent, nuanced and verified output. Essentially, AI is broken. It lies, it makes things up, it doesn't really understand - it's not conscious - it doesn't know what it is doing and doesn't understand reasoning it PERFORMS reasoning. There is a huge gap of epistemology that stops AI from being useful - one that can be patched with natural language instruction and vigilance - plus outside tools (Sympy, Numpy, Formal verification methods etc) - the key insight: When you clearly define what "right" is and what "wrong" means - the LLM can perform the task adequately - the general population assumes that because the AI responds eloquently and they understand what it is saying - that it understands nuance - that it has "common sense"- an assumption of intellect that is misinformed. AI "understands" the definition of a word - but until you articulate it in a prompt - a framework "You must tell me if my writing here is good, and by good I mean that, were it to be entered in a writing competition - one of a high standard - let us say X - it would likely win Gold - the definition of "good writing" is something that is well paced, and causes a visceral reaction in the reader etc" You literally make an entire "skill" file - by having an LLM research deeply the topic of "What makes good writing "good" - then you attach that file in a prompt and say "This file defines what is good writing, use it and the information therein to analyze my book and tell me how good it is - where it could be improved and any other advice that is NOT generic - but targeted and very well informed - based on the actual content of my book and the target demographic" See Felicity? One prompt "Tell me how good my writing is" You'll get a trash generic response - likely bad advice even. Another prompt with a smart method (Creating a skill file/report containing the definition of what "good writing" is deeply and eloquently articulated with examples and so on and techniques people use to judge writing quality and examples of the best and most popular books of all time - and everything people broke down to understand WHY those books did so well - attaching that file to the prompt with explicit instructions) The response: Absolute gold - on a level no one could likely recieve except by being expert writers a lot of money for deep eloquent advice - even with that - who will go page by page - sentence by sentence in your books and give you such deep advice? You gonna hire Stephen King, Branden Sanderson, Pierce Brown and a dozen other writers to give you deep pointers on every single line in your books? Yeah. No. But AI? AI will do whatever you want, as long as you want - pretty much for free. The method I described above is an example - deeply relevant to your own work Feliciity - to show what AI can be used for in a way that is relevant to you. That is what Ark is - it is my ability to articulate - to teach - to use AI concentrated in a methodology - automated as much as I can -- but with me at the helm - pointed in the direction of any problem - I find a way to fix it. Having ingenuity - knowing exactly what you would need to know to do something effectively and how you would go about doing that something - and providing that to the LLM - a mixture of Prompt engineering and Context engineering - so that the AI can do what it was made to do - assist you. Become a reasoning partner - a exocortex - an extension of your own cognition. Now take the method I just described and apply it to Patent law, to medicine, to relationship advice - anything. You'll get the same outcome. Output - results - from the AI that is highly accurate, highly valuable and highly effective in making YOU better at whatever it is you are trying to do. Hope that helps. ~Mo On Wed, Apr 29, 2026 at 5:36 PM JACK SARFATTI @icloud.com> wrote: Ask Mohamed I do not understand AI LLMs.jacksarfatti.academia.edu On Apr 29, 2026, at 2:32 PM, Felicity Harley wrote:I’d like to write something on ARK but I don’t quite understand it - if you have time to put down a few sentences I can further research it and check it out with Mohammad .On Apr 29, 2026, at 5:10 PM, JACK SARFATTI @icloud.com> wrote: Interesting that Sam Altman and Elon Musk suing each other in Oakland.The AI tech we have may put them both out of business. Too soon to tell.Also if my SO(2,4) G* >> G local gauge gravity extension of Einstein’s General Relativityworks that puts Space X out of business. x.com/i/grok/share/9… Technological Surprise x.com/i/grok/share/c… x.com/i/grok/share/5… x.com/i/grok/share/0… jacksarfatti.academia.edu Begin forwarded message:From: "Academia.edu" @academia-mail.com> Subject: You just got 106 views on "Tversky Neural Nets, Conscious AGI and Hawking's Mind of God" Date: April 29, 2026 at 1:44:37 AM PDTTo: jacksarfatti@gmail.com Hi Jack,Congratulations! You uploaded your paper 2 days ago and it is already gaining traction. Total views since upload: You got 106 views from France, Germany, the Netherlands, Saudi Arabia, the United States, Brazil, Nigeria, Spain, Greece, India, the Russian Federation, Italy, Israel, Canada, the United Kingdom, Ukraine, Kazakhstan, Taiwan, Argentina, Belgium, Hong Kong, Algeria, and China on "Tversky Neural Nets, Conscious AGI and Hawking's Mind of God". Upload Another Paper Thanks, The Academia.edu Team You can update your preferences or unsubscribe. Academia.edu, 580 California St., Suite 400, San Francisco, CA, 94104