Mustafa

The pursuit of "being cracked"

@m_goes_distance that's some night city shit!

careers left in the singularity: - niche podcaster - sex machine operator - farmer - philosopher - time machine operator - human verification specialist - blood boy - trad wife - claude operator - cult leader - shitpoaster - peptide vendor - lover/warrior/magician - biohacker storyteller - monk - personality designer - afterlife curator - sec of state what else?

is a huge, high roofed, white floored warehouse the cure to male depression?

the bottleneck in achieving AGI isn't compute or algorithms; but the probabilistic approach on deterministic machines.

maxwell’s equations define how fields behave, not just forces they replace “action at a distance” with local interactions in space and time four constraints: → gauss (electric) ∇·E = ρ/ε₀ charge density creates divergence in the electric field → gauss (magnetic) ∇·B = 0 no magnetic sources, only loops → faraday ∇×E = −∂B/∂t changing magnetic field induces circulation in E → ampere – maxwell ∇×B = μ₀J + μ₀ε₀ ∂E/∂t currents and changing E generate B structure: → divergence ties fields to sources → curl ties fields to dynamics key move: → maxwell added the displacement current term (∂E/∂t) without it, symmetry breaks and waves don’t exist result: → combine curls → wave equations → fields sustain each other and propagate c = 1 / √(μ₀ε₀) light = electromagnetic wave implications: → no medium required → information carried by field oscillations in engineering: → antennas radiate by accelerating charges → circuits leak energy as radiation at high frequency in computation: → full-wave solvers approximate these PDEs interpretation: → not four separate laws → a coupled system defining allowable field evolution maxwell’s equations are a constraint system on reality everything electromagnetic must satisfy them

convolutional neural networks are built for spatial intelligence regular networks ignore structure CNNs exploit it images aren’t flat vectors they have locality, patterns, hierarchy what a CNN does: → slides filters across input → detects local features → builds complexity layer by layer core components: → convolution small kernel scans the input learns edges, textures, patterns → activation non-linearity so it doesn’t stay linear → pooling compresses information keeps what matters, drops redundancy → stacking layers edges → shapes → objects why it works: → weight sharing reduces parameters → locality preserves spatial relationships → hierarchy builds abstraction in practice: → image classification → object detection → medical imaging → video understanding key idea: → same pattern, different location = same meaning limitations: → struggles with long-range dependencies → fixed receptive fields unless extended evolution: → CNNs dominated vision → now combined or replaced in parts by transformers CNNs are not just models they’re engineered bias toward how the visual world is structured

study materials science. every physical system is limited by its materials. • strength. • weight. • conductivity. • temperature limits. • fatigue. • corrosion. these are not secondary details. they define what is possible. materials science links structure to behavior: atomic arrangement → microstructure → properties → performance. change the structure, and everything changes. learn the fundamentals: • Crystal Structure • Dislocation • Phase Diagram • Stress–Strain Curve this is how engineers design: • aerospace alloys • semiconductor devices • battery systems • composites • high-temperature materials you are not just choosing materials. you are engineering matter itself.

curiosity maxxing is the only way out of average thinking most people consume few actually investigate they stop at surface-level answers you go one layer deeper every time what it looks like: → you don’t accept explanations, you break them apart → you chase “why” until it hits first principles → you connect domains that shouldn’t connect pattern: → question → dig → model → rebuild why it matters: → curiosity compounds into understanding → understanding compounds into leverage in engineering: → you don’t just use tools, you understand how they’re built in AI: → you don’t just train models, you question the assumptions in life: → you stop reacting, you start seeing structure constraints: → curiosity without direction = noise → curiosity with systems = power most people wait to be taught curiosity maxxing forces you to discover that’s the difference between knowing and actually understanding

AI without neuroscience is blind pattern fitting you’re copying outputs without understanding the system that inspired it study neuroscience what you unlock: → how real intelligence encodes information → how learning actually happens (plasticity, not just gradients) → how memory is structured (not just vectors, but dynamics) key parallels: → neurons ≠ artificial neurons real ones spike, adapt, rewire → learning ≠ backprop brains use local rules, timing, chemistry → intelligence ≠ scale it’s efficiency, structure, constraints why it matters: in deep learning: → architectures come from biology (CNNs, attention roots in perception) in AGI: → you need models of cognition, not just bigger models in robotics: → perception + action loops are biological problems in optimization: → energy-based views come straight from brain dynamics what to study: → spiking neural networks → synaptic plasticity (hebbian learning, STDP) → predictive coding → cortical hierarchies direction: → current AI = approximation → neuroscience = source code if you ignore the source you plateau fast

the fourier transform is how you see structure inside signals time domain lies to you frequency domain exposes it any signal → sum of pure oscillations sines and cosines are the basis what it does: → decomposes a function into frequencies → tells you “what frequencies exist” and “how strong they are” continuous form: F(ω) = ∫ f(t) e^(−iωt) dt discrete world: → DFT / FFT → turns sampled data into usable spectra why it matters: in engineering: → filter noise → design communication systems → analyze vibrations, audio, RF in physics: → waves, heat, quantum states → solutions become algebra in frequency space in computing: → image compression (JPEG) → signal processing pipelines → fast convolution via frequency multiplication key idea: → convolution in time = multiplication in frequency → complexity collapses interpretation: → sharp spikes → dominant frequencies → spread spectrum → complex / noisy signal the fourier transform is not about math it’s about changing perspective to make systems tractable

learn chinese.

the jacobian matrix is how multivariable systems actually move you don’t deal with one variable anymore you deal with transformations input vector → output vector the jacobian captures how every input dimension affects every output dimension what it is: → a matrix of partial derivatives → each row = one output function → each column = one input variable J(i,j) = ∂f_i / ∂x_j why it matters: → it’s the local linear approximation of a nonlinear system → it tells you how small changes propagate → it converts messy systems into something you can compute in physics: → coordinate transformations → velocity mappings → change of variables in integrals in robotics: → maps joint velocities → end-effector velocity → singularities show up when the jacobian collapses in optimization / ML: → gradient flow through layers → backprop is chained jacobians interpretation: → determinant ≠ 0 → transformation is locally invertible → determinant = 0 → information collapse the jacobian is not theory it’s the interface between geometry and computation

@Currentreport1 what happened to the .07% who voted against him?

BREAKING: Kim Jong-un has won North Korea's parliamentary elections with 99.93% of the vote.

the geodesic equation is how nature finds the “straightest” path. not straight in the usual sense; straight within a curved space. what it means: • shortest/least-action path → objects move along paths that minimize distance or energy • curved space → in flat space it’s a line, in curved space it bends • gravity as geometry → mass curves spacetime, objects just follow that curvature • no force needed → motion under gravity is just following a geodesic examples: • planets orbiting stars → not “pulled,” just following curved spacetime • light bending near massive objects → even photons follow geodesics • great circles on earth → shortest path between two points on a sphere why it matters: physics stops being about forces pushing things around and becomes about geometry telling things where to go the geodesic equation is the rule that turns curved space into motion.

NVIDIA Headquarters - Santa Clara

the art of 3D printing

@indiranegi fair enoughh

@oprydai The fact that my friends have labs like this so I can just go there 😊. Have a small version myself

whats stopping you from building a home-lab like this?

any sufficiently advanced engineering is indistinguishable from art

study Materials Science. every physical system is limited by its materials. strength. weight. conductivity. temperature limits. fatigue. corrosion. these are not secondary details. they define what is possible. materials science links structure to behavior: atomic arrangement → microstructure → properties → performance. change the structure, and everything changes. learn the fundamentals: • Crystal Structure • Dislocation • Phase Diagram • Stress–Strain Curve this is how engineers design: • aerospace alloys • semiconductor devices • battery systems • composites • high-temperature materials you are not just choosing materials. you are engineering matter itself.

@TensorTonic because of the smooth approximation near zero, while ReLU cuts off hard, also the gradient flow in DNs.

Can you explain why ReLU kills gradients, why GELU is in every transformer, why Softmax turns logits into probabilities? > Sigmoid > ReLU > Tanh > Softmax > LeakyReLU > GELU > Swish > ELU > SELU Every activation function you'll ever need, explained by implemention. Practice all of them on TensorTonic: tensortonic.com