https://www.neuralialabs.com/en/videos/gradient-checking 2026-10-02T04:02:05.050Z https://storage.neuralialabs.com/social/4sCaJOMdolOC4HitPLt5/story/cover.87c8eea3.jpg Gradient checking: catching the silent bug in AI training math — AI Concepts #032 A hand-derived slope formula can be quietly wrong with no error message. Learn how gradient checking measures slopes directly and exposes a one-character typo. https://storage.neuralialabs.com/social/4sCaJOMdolOC4HitPLt5/story/video.f88be094.mp4 23 2026-10-02T04:02:05.050Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/momentum 2026-10-01T22:01:12.365Z https://storage.neuralialabs.com/social/1Sa5I6LCjphWhPtGTJPD/story/cover.9575a92d.jpg Momentum: how remembering past steps stops AI training zigzagging — AI Concepts #031 In a narrow valley, training zigzags and wastes most of each step. Learn how momentum cancels the zigzag, turning 79,513 steps into 461 with two lines of code. https://storage.neuralialabs.com/social/1Sa5I6LCjphWhPtGTJPD/story/video.36bfec58.mp4 24 2026-10-01T22:01:12.365Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/mini-batch-sgd 2026-10-01T16:00:06.163Z https://storage.neuralialabs.com/social/ZaepJXPpnxPaWGLyUfQ0/story/cover.709309dd.jpg Mini-batches: why AI learns faster from a random handful of data — AI Concepts #030 Why training on a random handful of 32 examples beats using the whole dataset at every step: noisier steps, but 219 times less work to reach the same place. https://storage.neuralialabs.com/social/ZaepJXPpnxPaWGLyUfQ0/story/video.0aa3d49f.mp4 25 2026-10-01T16:00:06.163Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/saddle-point 2026-10-01T10:04:06.664Z https://storage.neuralialabs.com/social/vtP9QkgFS6OIMAfzOka0/story/cover.ed13f32e.jpg Saddle points: where AI training looks stuck but has a way down — AI Concepts #029 A flat spot is not always a valley. Learn what a saddle point is, why gradient descent cannot tell it from a minimum, and why big networks are full of them. https://storage.neuralialabs.com/social/vtP9QkgFS6OIMAfzOka0/story/video.68960e55.mp4 23 2026-10-01T10:04:06.664Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/local-minima 2026-10-01T04:03:06.758Z https://storage.neuralialabs.com/social/aukceW7tLTqhkyzPgPGb/story/cover.8535d47b.jpg Local minima: why AI training can settle for second best — AI Concepts #028 Walking downhill ends in the valley you start in, not the deepest. Learn what a local minimum is, how the start decides it, and why big networks worry less. https://storage.neuralialabs.com/social/aukceW7tLTqhkyzPgPGb/story/video.2e6d6934.mp4 24 2026-10-01T04:03:06.758Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/condition-number 2026-09-30T22:02:06.700Z https://storage.neuralialabs.com/social/9MNUmLU7VsWM0MjAUpeh/story/cover.a661c4b6.jpg Condition number: why a stretched valley makes AI training crawl — AI Concepts #027 A long, narrow error valley forces tiny steps and slow learning. See what the condition number measures and why centring data cut 79,513 steps to 10. https://storage.neuralialabs.com/social/9MNUmLU7VsWM0MjAUpeh/story/video.a7865493.mp4 25 2026-09-30T22:02:06.700Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/learning-rate 2026-09-30T16:01:12.579Z https://storage.neuralialabs.com/social/Kuhl2nZrdZ0XOxKDYSaS/story/cover.76d532cb.jpg Learning rate: the step size with a ceiling you can calculate — AI Concepts #026 Too small is slow, just right lands fast, and past an exact limit training blows up. Learn what the learning rate is and how its ceiling is calculated. https://storage.neuralialabs.com/social/Kuhl2nZrdZ0XOxKDYSaS/story/video.4b52f00f.mp4 26 2026-09-30T16:01:12.579Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/gradient-descent 2026-09-30T10:01:05.233Z https://storage.neuralialabs.com/social/2mGqd1wfl7zjMZx88Hui/story/cover.919b9255.jpg Gradient descent: how AI finds its way downhill in small steps — AI Concepts #025 Instead of trying every setting, an AI steps against the slope of its error and repeats. Learn how gradient descent works, and why too big a step backfires. https://storage.neuralialabs.com/social/2mGqd1wfl7zjMZx88Hui/story/video.3cc02281.mp4 24 2026-09-30T10:01:05.233Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/gradient 2026-09-30T04:00:39.818Z https://storage.neuralialabs.com/social/4CtTkFde9Jv5MF2Ren9p/story/cover.7f2d3aae.jpg Gradient: the arrow that points uphill, and why AI walks the other way — AI Concepts #024 The gradient combines the slope of every setting into one arrow. Learn why it points where the error rises fastest, and why AI steps the opposite way. https://storage.neuralialabs.com/social/4CtTkFde9Jv5MF2Ren9p/story/video.996fb1df.mp4 25 2026-09-30T04:00:39.818Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/chain-rule 2026-09-29T22:04:06.131Z https://storage.neuralialabs.com/social/RRC5QrC5ZPfyZUhtz7Di/story/cover.648de0f8.jpg Chain rule: how AI traces the effect of every layer — AI Concepts #023 When one change drives another, which drives a third, the rates multiply. Learn the chain rule, the idea that lets AI trace how each layer shapes the result. https://storage.neuralialabs.com/social/RRC5QrC5ZPfyZUhtz7Di/story/video.f79d48c8.mp4 23 2026-09-29T22:04:06.131Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/derivative 2026-09-29T16:03:14.184Z https://storage.neuralialabs.com/social/jU1SCYoJbzJNuhYxDinv/story/cover.60cad28e.jpg Derivative: how much the result moves when you nudge one input — AI Concepts #022 A derivative measures how fast a result changes when you nudge an input. Learn how AI uses it, and why a nudge that is too tiny gives a worse answer. https://storage.neuralialabs.com/social/jU1SCYoJbzJNuhYxDinv/story/video.322dcdb9.mp4 24 2026-09-29T16:03:14.184Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/grid-search 2026-09-29T10:02:08.464Z https://storage.neuralialabs.com/social/CQVxgwwqxHUj27zVGblN/story/cover.2e886c63.jpg Grid search: why you cannot train an AI by trying every setting — AI Concepts #021 Trying every combination finds the best settings for a tiny model. Here is why each extra setting multiplies the work until search becomes impossible. https://storage.neuralialabs.com/social/CQVxgwwqxHUj27zVGblN/story/video.f2d3990a.mp4 25 2026-09-29T10:02:08.464Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/robust-loss 2026-09-29T04:01:05.181Z https://storage.neuralialabs.com/social/xtjreS2Su2KhfukdySpi/story/cover.29a5014c.jpg Robust loss: stop one wild reading from wrecking your forecast — AI Concepts #020 One bad measurement can push a standard fit hours off. Learn how a robust loss still counts it without letting it decide, with a worked factory example. https://storage.neuralialabs.com/social/xtjreS2Su2KhfukdySpi/story/video.e50f08a3.mp4 26 2026-09-29T04:01:05.181Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/heavy-tailed-noise 2026-09-28T22:00:07.482Z https://storage.neuralialabs.com/social/8fYEBKZKbuem8rFeked4/story/cover.d965a495.jpg Heavy tails: when rare, huge errors break the usual average — AI Concepts #019 Some errors are tiny almost always, then suddenly enormous. Learn what heavy-tailed noise is, how to spot it, and why the usual averages stop working. https://storage.neuralialabs.com/social/8fYEBKZKbuem8rFeked4/story/video.d00867a9.mp4 25 2026-09-28T22:00:07.482Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/mean-squared-error 2026-09-28T18:04:05.619Z https://storage.neuralialabs.com/social/c9PqNFot6ICxqnxtaBYy/story/cover.91f9af7c.jpg Mean squared error: the hidden bet behind squaring your mistakes — AI Concepts #018 Squaring errors is not just a habit. Learn what mean squared error quietly assumes about your data, when that bet pays off, and when it misleads you. https://storage.neuralialabs.com/social/c9PqNFot6ICxqnxtaBYy/story/video.4e7ab458.mp4 25 2026-09-28T18:04:05.619Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/logsumexp-trick 2026-09-28T12:03:10.207Z https://storage.neuralialabs.com/social/jzBvHY3sY3Q9DsWn7i2U/story/cover.e61c158d.jpg The LogSumExp trick: one subtraction that stops AI math overflowing — AI Concepts #017 Some AI formulas produce numbers too big for a computer to hold. Learn how subtracting the largest value first gives the exact same answer, with no crash. https://storage.neuralialabs.com/social/jzBvHY3sY3Q9DsWn7i2U/story/video.8525f772.mp4 23 2026-09-28T12:03:10.207Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/floating-point-rounding-error 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/8tN8y7vWHODs5FcHvuWu/story/cover.6d181fcc.jpg Rounding error: why adding the same numbers can give two answers — AI Concepts #016 Add a million numbers in two different orders and a computer returns two totals, both wrong. Learn why every sum rounds, and why even chatbots feel it. https://storage.neuralialabs.com/social/8tN8y7vWHODs5FcHvuWu/story/video.a76e45c6.mp4 24 2026-09-28T05:02:30.271Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/floating-point-numbers 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/t4cTCgqaawtiu77b3mFJ/story/cover.5929ce26.jpg Floating-point numbers: why your computer cannot hold every number — AI Concepts #015 A computer number has a fixed budget of bits, split between detail and range. Learn how that split works and why two 16-bit formats fail so differently. https://storage.neuralialabs.com/social/t4cTCgqaawtiu77b3mFJ/story/video.1071749b.mp4 26 2026-09-27T18:02:21.209Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/negative-log-likelihood 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/5EyzzidJV55P7w8QoqRF/story/cover.0faca505.jpg Negative log-likelihood: turning a fragile product into a calm sum — AI Concepts #014 Multiplying thousands of chances breaks a computer. Taking logarithms turns the product into a sum with the same winner, and squared error falls out of it. https://storage.neuralialabs.com/social/5EyzzidJV55P7w8QoqRF/story/video.0f51e411.mp4 25 2026-09-27T12:01:13.574Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/maximum-likelihood 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/imiK5jBZj3oqIoY3rsb5/story/cover.f616a231.jpg Maximum likelihood: the answer that makes your data least surprising — AI Concepts #013 Maximum likelihood picks the answer under which what you actually observed looks least surprising. A plain guide with a coin, a bell curve and real numbers. https://storage.neuralialabs.com/social/imiK5jBZj3oqIoY3rsb5/story/video.39ea0663.mp4 25 2026-09-27T05:01:13.388Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/bayes-rule 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/OYrzDIEmbRO8pFt0ZwU7/story/cover.1118cea8.jpg Bayes' rule: reasoning backwards from a clue to its most likely cause — AI Concepts #012 Rain wets streets, but does a wet street mean rain? Bayes' rule flips the question and weighs how common each cause is. A plain guide with one worked example. https://storage.neuralialabs.com/social/OYrzDIEmbRO8pFt0ZwU7/story/video.589997d8.mp4 24 2026-09-26T18:00:58.779Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/chain-rule-of-probability 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/kKvGtiptcqLPYYqavxoG/story/cover.4b617a4e.jpg Chain rule of probability: how AI scores a sentence word by word — AI Concepts #011 The chance of a whole sentence is the chance of each word given the words before it, multiplied together. Why this exact rule sits under language models. https://storage.neuralialabs.com/social/kKvGtiptcqLPYYqavxoG/story/video.b0e56ae4.mp4 24 2026-09-26T12:04:20.335Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/loss-landscape 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/feLHGsq64jRIhjd4gdDw/story/cover.7569a511.jpg Loss landscape: training an AI means finding the bottom of a valley — AI Concepts #010 Score a model at every possible setting and its errors form a landscape. Here is why a smooth valley tells training which way is downhill, and a staircase does not. https://storage.neuralialabs.com/social/feLHGsq64jRIhjd4gdDw/story/video.1ac3d785.mp4 23 2026-09-26T05:04:19.399Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/loss-function 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/dUP22LSZ35PcZ0NNRqaJ/story/cover.b82420ee.jpg Loss function: the scoring rule that decides which answer wins — AI Concepts #009 A loss function turns a model's mistakes into one score. Swap the rule and a different answer wins, which is why choosing it means defining the problem. https://storage.neuralialabs.com/social/dUP22LSZ35PcZ0NNRqaJ/story/video.4190f47a.mp4 24 2026-09-25T18:03:23.364Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/feature-engineering 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/HlKkScZKA7yo1kZQwqjv/story/cover.8d701678.jpg Feature engineering: the human choice that can beat the algorithm — AI Concepts #008 Before a model learns anything, someone decides what to measure. Learn what feature engineering is, why it can outweigh the algorithm, and what replaced it. https://storage.neuralialabs.com/social/HlKkScZKA7yo1kZQwqjv/story/video.abe1b5b9.mp4 22 2026-09-25T12:01:11.177Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/xor-problem 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/rb3jOLFRIlXXH3NersKq/story/cover.339974c8.jpg The XOR problem: four dots no straight line can separate — AI Concepts #007 Why a single perceptron, one of the simplest learning programs, can never learn "exclusive or", what it does when trained anyway, and what finally fixed it. https://storage.neuralialabs.com/social/rb3jOLFRIlXXH3NersKq/story/video.898acfa9.mp4 25 2026-09-25T05:00:31.988Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/input-normalization 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/3PpVRmFL3XqWM1wR6vrh/story/cover.8e2aecff.jpg Input normalization: three lines of code, 6,000 times faster training — AI Concepts #006 Subtracting the average from your data let the same simple model train in 2 passes instead of 11,976. Learn why moving the zero point helps so much. https://storage.neuralialabs.com/social/3PpVRmFL3XqWM1wR6vrh/story/video.300efff4.mp4 23 2026-09-24T18:04:17.757Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/perceptron-convergence-theorem 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/KPbEbaOROxAYZVXsp8mJ/story/cover.4c24a251.jpg Convergence theorem: guaranteed to finish, silent about when — AI Concepts #005 A classic theorem promises a simple AI learner will find its answer, if one exists. Learn why it says nothing about time, and what decides how long it takes. https://storage.neuralialabs.com/social/KPbEbaOROxAYZVXsp8mJ/story/video.8a51fd61.mp4 25 2026-09-24T12:04:16.746Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/perceptron-learning-rule 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/PtTGZVS14uzy892LTKD1/story/cover.b5801e97.jpg Learning from mistakes: the perceptron rule that only moves when wrong — AI Concepts #004 How a perceptron learns: change nothing when right, nudge toward the answer when wrong. Why that nudge is justified, and why it needs no calculus at all. https://storage.neuralialabs.com/social/PtTGZVS14uzy892LTKD1/story/video.ca3099c4.mp4 23 2026-09-24T05:04:15.429Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/dot-product 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/LFScNSsHG3DihjakJRNC/story/cover.95863643.jpg Dot product: the multiply-and-add step at the heart of AI's math — AI Concepts #003 The dot product multiplies two lists of numbers pair by pair and adds them up. See it on a shopping receipt first, then inside every neural network layer. https://storage.neuralialabs.com/social/LFScNSsHG3DihjakJRNC/story/video.55aac1bc.mp4 23 2026-09-23T18:04:14.558Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/decision-boundary 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/HlcQHNKCUuth2fJoauD5/story/cover.0a29f9d9.jpg Decision boundary: the line a simple model draws between yes and no — AI Concepts #002 A decision boundary is where a simple AI model cannot make up its mind. See how the weights set the line's angle, the bias slides it, and training moves it. https://storage.neuralialabs.com/social/HlcQHNKCUuth2fJoauD5/story/video.5c4fdaeb.mp4 22 2026-09-23T12:04:10.590Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/perceptron 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/TxTERoReVwQJG5wn3hYo/story/cover.516d0283.jpg Perceptron: the yes-or-no model every neural network grew from — AI Concepts #001 What a perceptron is, how it turns a few measurements into a yes or a no, and how the same data took 29,870 corrections raw but just one once centred. https://storage.neuralialabs.com/social/TxTERoReVwQJG5wn3hYo/story/video.72877d93.mp4 47 2026-09-23T05:03:14.755Z AI Concepts yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/embeddings 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/fmnAuF0P9xCudSF75k97/story/cover.afb10bfd.jpg Embeddings: why a puppy lands closer to a dog than to a pizza — AI 101 #06 Embeddings turn words into points so related words end up near each other. See how that is learned, and why the famous king-to-queen trick is weaker than it looks. https://storage.neuralialabs.com/social/fmnAuF0P9xCudSF75k97/story/video.2b6ec970.mp4 20 2026-09-22T18:02:16.019Z AI 101 yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/tokens 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/lhv5iLhZk3Re62PnOyLJ/story/cover.739ca932.jpg Tokens: how AI chops your words into numbered pieces — AI 101 #05 Chatbots never see your words, only numbered pieces called tokens. Learn how text gets split, why long words break apart, and why it changes what you pay. https://storage.neuralialabs.com/social/lhv5iLhZk3Re62PnOyLJ/story/video.df6c95d1.mp4 20 2026-09-22T12:02:14.676Z AI 101 yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/how-ai-learns 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/AZUlag3ljW3OG0FL0yKK/story/cover.112f7cb9.jpg How AI learns: guess, measure the error, nudge, repeat — AI 101 #04 AI learns by guessing, measuring how wrong it was and nudging its internal knobs, over and over. Here is that loop in plain words, with a guitar example. https://storage.neuralialabs.com/social/AZUlag3ljW3OG0FL0yKK/story/video.b0628654.mp4 20 2026-09-22T05:02:15.327Z AI 101 yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/neural-networks 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/VGC9hQ0V0OPQ0yaNQFJ0/story/cover.b23b25f3.jpg Neural networks: layers of tiny calculators passing signals on — AI 101 #03 A neural network is layers of simple neurons passing numbers forward. See what one neuron really computes, and why layers solve what a single line cannot. https://storage.neuralialabs.com/social/VGC9hQ0V0OPQ0yaNQFJ0/story/video.35604647.mp4 20 2026-09-21T17:05:57.905Z AI 101 yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/machine-learning 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/LUn5aOGflHJEkVZugtAB/story/cover.4e19ed8c.jpg Machine learning: how a computer draws its own dividing line — AI 101 #02 Nobody writes the rule: the computer draws a line between examples, checks its mistakes and nudges the line, then uses it to sort new cases. Here is how. https://storage.neuralialabs.com/social/LUn5aOGflHJEkVZugtAB/story/video.d5c8b1c6.mp4 20 2026-09-21T09:22:22.124Z AI 101 yes NeuraLIA Labs https://www.neuralialabs.com/en/videos/what-is-ai 2026-09-28T06:16:17.579Z https://storage.neuralialabs.com/social/5G2REV2S8O0Wn1aSsVKH/story/cover.0a00dcab.jpg What is AI? Software that learns its own rules from examples — AI 101 #01 Regular software follows rules that someone wrote. Modern AI learns them from examples instead. Here is what that swap means, with a handwriting example. https://storage.neuralialabs.com/social/5G2REV2S8O0Wn1aSsVKH/story/video.d46cda71.mp4 20 2026-09-21T07:02:18.277Z AI 101 yes NeuraLIA Labs