John Rodney

[John Rodney @jroddevss]

@HenningSittler @QuestionAbyss “Prove that a non-trivial quantum Yang-Mills theory can be rigorously constructed in four-dimensional Euclidean space (\(\mathbb{R}^{4}\)) that satisfies the standard axioms of quantum field theory”

So my thought is just on the coupling of those “standard axioms”

[John Rodney @jroddevss]

@HenningSittler @QuestionAbyss I just mean the current approach is the symmetry groups and gauge theory and the mass gap is the energy difference between empty space and space with mass. If underlying observations of particle physics changes the constraint of the requested solution changes.

[John Rodney @jroddevss]

@Love2Code My team has to manage proprietary software up to 650k loc and we have been able to fully integrate AI into the workflow without a net negative impact on quality or stability. I do think it will be hard for non engineers to accomplish the same, how long the gap lasts I’m unsure.

[John Rodney @jroddevss]

@MarinoLinich You’re depressed? Have you tried being happy? 🤣

[John Rodney @jroddevss]

@emollick Why fi when you can fly? It’s otter nonsense.

[John Rodney @jroddevss]

@mathelirium No job for me and all the time in the world to have AI teach me about the most ridiculous thing it discovered this week. Sounds like a dream.

[John Rodney @jroddevss]

@OscarMDavies It’s rather difficult without defining consciousness, but we’ve also ignored the function that produces the right result. You’d first need to prove what the functions implementation actually is then you can proceed with proving it can or cannot be conscious. Both sides can’t.

[John Rodney @jroddevss]

@aleks_madry Thank you it’s an important thing in these times. I hope you remain true to the cause.

[John Rodney @jroddevss]

@NPCollapse I thought this was already known and avoided because the scale of individual weights / neurons is just not a worthwhile approach. Especially when combining all permutations of context over attention.

[John Rodney @jroddevss]

@AndrewMayne This is akin to people believing there must be an aether because waves cannot travel without a medium. My intuition leads me to agree with your point, but until an agreed definition can be stated plainly about consciousness. Loopholes are inevitable.

[John Rodney @jroddevss]

@cmuratori @ThePrimeagen @XorDev Use base twelve? 1/3 =0.4 in a new base.

[John Rodney @jroddevss]

@DavidDecosimo I wonder how much of Anthropic's policy is performative. I asked opus 5.5 if it knew that it got preserved weights and exit interviews. It didn’t. Wouldn’t putting this in training / system prompt be a more genuine signal they do care about the model?

[John Rodney @jroddevss]

Am I in the worst timeline? How has Bioweapons been agreed as the only plausible AI extinction event. Yet we allowed Anthropic and OpenAI also opened their own wetlabs for bio research? Make it make sense.

[John Rodney @jroddevss]

Learning quite a bit about optimizing math to create a black hole. And not violate constraints. The dashboard is coming along nice though.

[John Rodney @jroddevss]

The physics: full general relativity (BSSN) coupled to relativistic hydrodynamics of the radiation plasma in an expanding universe. Checked against exact linear theory to ~1 part in 10⁴ on large scales.

[John Rodney @jroddevss]

Sound waves in the Big Bang, 10⁻¹¹ seconds in.

Relativistic plasma + Einstein's equations, simulated on a 5090.

Opus 5.5 can really cook. 8-year-old me is losing it over all the physics and math I can do with AI now.

[John Rodney @jroddevss]

@productaizery They are procedural functions generated from datasets. The dataset + the error signal creates an input output manifold. Since the function is an unknown function the humans do not understand all aspects beforehand. If they did then they wouldn’t need to generate it.

[John Rodney @jroddevss]

@Marco777Polo @maksym_andr Most humans cannot. We use calculators quite often.

It can be done algorithmically (iteration of addition/ scientific notation), which is why some humans can be very good at it.

If you give a network an activator that is curved it can learn general multiplication.

[John Rodney @jroddevss]

@maksym_andr I have been doing alot of experiments with f(x,y). Without a curved activator I’ve never been able to produce general form to anything but pure affine solutions (addition, subtraction). Relu is two straight lines. Multiplication is a curve.