1763Mathematics
Bayes' Theorem: The Equation That Came Back from the Dead
Thomas Bayes never published his most famous result. A friend found it among his papers after he died.
The History of AI
Every algorithm running in production today has an origin story. A game theorist dividing poker winnings fairly. A mathematician proving free will doesn't exist in Russian poetry. A Navy-funded machine promised to walk, talk, and be conscious. These are those stories.
1805Mathematics
Least Squares: The Statistical War That Shaped Every Regression
Gauss claimed he invented it. Legendre published it first. The dispute never fully resolved, and it's still the most widely used statistical technique in history.
1901Algorithms
PCA: The Algorithm That Finds the Shape Hidden in Your Data
Karl Pearson invented PCA to draw the best line through a cloud of points. It became the most widely used dimensionality reduction technique in science.
1906Mathematics
Markov Chains: From a Poetry Dispute to Language Models
Markov Chains were invented to prove free will doesn't exist in Russian poetry.
1925People
The p-value: How a Convenient Threshold Became Science's Gatekeeper
Ronald Fisher suggested p < 0.05 as a rule of thumb in 1925. He never meant it to become the global standard for what counts as true.
1948People
Cybernetics: The Science That Almost Became AI
Norbert Wiener invented cybernetics to understand how brains, machines, and missiles aim at targets.
1950People
The Turing Test: A Question That Defined a Century
Turing's original test involved a human, a machine, and a gender-guessing game.
1953Mathematics
SHAP: From Game Theory Tables to Model Explainability
SHAP values come from a 1953 paper on dividing poker winnings fairly.
1955Moments
The Logic Theorist: The World's First AI Program
The Logic Theorist proved 38 theorems from Principia Mathematica, and the authors tried to submit it as a co-author.
1956Moments
Dartmouth 1956: The Summer That Named a Field
"Artificial Intelligence" was coined at a summer workshop with 10 attendees.
1958Models
The Perceptron: A Machine That Would Think
The Perceptron promised to walk, talk, and be conscious within a decade.
1966People
ELIZA: The Chatbot That Unsettled Its Own Creator
ELIZA's creator was disturbed that people genuinely opened up to his chatbot.
1967Algorithms
K-Means: The Forgotten Algorithm That Clusters Everything
K-means was published in an IBM report that almost no one read for 20 years.
1969Moments
The XOR Problem: How a Single Function Froze a Field
Two mathematicians nearly killed neural networks for 15 years.
1976Moments
Expert Systems: The AI Winter They Built
Expert systems promised to bottle human expertise. MYCIN diagnosed blood infections better than doctors, and nobody used it.
1979Algorithms
The Bootstrap: Pulling Statistical Inference from the Data Itself
Bradley Efron realised you could estimate statistical uncertainty by sampling your own data, without knowing anything about the underlying distribution.
1982Models
Hopfield Networks: Memory as Energy
John Hopfield showed that a network of neurons could function as an associative memorylike a brain that recalls the whole from a fragment.
1986Algorithms
Backpropagation: The Algorithm That Would Not Be Published
Backprop was rejected by journals for years before it changed everything.
1989Models
LeNet: The Convolutional Net That Read Your Cheques
LeCun's convolutional network read handwritten cheques for AT&T10% of all cheques cleared in the US by 1998.
1991Algorithms
The Vanishing Gradient: Why Deep Networks Couldn't Learn, Until LSTM
The vanishing gradient problem sat unsolved for 10 years before LSTM fixed it.
1991Algorithms
Mixture of Experts: The Forgotten Architecture Behind the Frontier
Mixture of Experts was invented in 1991 and largely forgotten. Then it became the architecture reportedly used in GPT-4.
1992Models
TD-Gammon: The Neural Network That Taught Itself to Play
TD-Gammon taught itself backgammon by playing 1.5 million gamesalone.
1992Algorithms
Support Vector Machines: The Maximum Margin Classifier
Support Vector Machines dominated machine learning for a decadebuilt on a 1963 idea about the widest possible gap between classes.
1996Datasets
Adult Income: The Dataset That Exposed ML's Fairness Problem
The Adult Income dataset was extracted from a census to study poverty, and accidentally became the standard test for algorithmic fairness.
1997Mathematics
No Free Lunch: Why There's No Best Algorithm
The No Free Lunch theorem proved that no algorithm is universally better than any other, averaged over all possible problems.
1997Datasets
California Housing: The Benchmark That Replaced a Biased One
California Housing replaced the Boston dataset after researchers found Boston embedded racial data in a feature called "lower status." California became the clean regression standard.
1997Moments
Deep Blue vs. Kasparov: The Day a Machine Beat a World Champion
In May 1997, a machine without intuition beat the best chess player alive, and proved that brute-force search can look exactly like genius from the outside.
1998Models
MNIST: The Dataset That Trained a Generation
MNIST started as a post office problem: sorting zip codes by hand.
1998Algorithms
PageRank: When Link Structure Became a Markov Chain
PageRank treated the web as a Markov chain, and the math that ranked websites is now the backbone of Google's $2 trillion empire.
2001Algorithms
Random Forests: Wisdom of the Crowd, One Tree at a Time
Random Forests exist because one scientist was frustrated by noisy single trees.
2006Algorithms
Monte Carlo Tree Search: Random Sampling Meets Strategic Depth
Monte Carlo Tree Search made computers competitive at Go a decade before AlphaGo, by simulating thousands of random games.
2006Moments
CUDA: The Accidental Infrastructure of the AI Age
NVIDIA built CUDA to sell more GPUs for gaming. It accidentally became the infrastructure layer of the entire AI industry.
2009Datasets
CIFAR-10: The Benchmark Between MNIST and ImageNet
CIFAR-10 was assembled by Alex Krizhevsky in Hinton's lab, and the techniques he developed to crack it became AlexNet.
2011Datasets
IMDB Sentiment: The Dataset That Brought NLP into the Deep Learning Era
Andrew Maas scraped 50,000 IMDB reviews to study word vectors for sentiment, and created NLP's first large-scale transfer learning benchmark.
2012Moments
ImageNet: The Dataset That Triggered the Deep Learning Era
ImageNet was built by crowdsourcing 14 million images: labelled for $0.01 each.
2012Algorithms
Dropout: The Regularisation Trick Born in a Bank
Dropout was inspired by how banks prevent fraudnot letting tellers know everything.
2012Datasets
Titanic: The Dataset That Taught a Million People Data Science
Kaggle launched with the Titanic dataset and accidentally created the on-ramp for a generation of data scientists.
2012Models
AlexNet: The Network That Woke Up Deep Learning
In September 2012, two grad students and their professor submitted a network to an obscure vision contest and ended the hand-engineered-features era of AI in a single afternoon.
2013Models
Word2Vec: The Geometry of Language
Word2Vec proved that king − man + woman ≈ queen: geometry in meaning.
2013Models
DQN: Teaching a Network to Play Atari from Pixels
DeepMind's DQN learned to play 49 Atari games from raw pixels, and beat humans on 29 of them.
2013Datasets
Credit Card Fraud: The Dataset That Taught ML to Handle Imbalance
A European credit card dataset with 492 frauds in 284,807 transactions taught the field how to handle extreme class imbalance.
2014Algorithms
Attention: The Mechanism That Rewired Everything
Attention was invented to help machines translate French novels.
2014Models
GANs: The Art of Machine Deception
Ian Goodfellow invented GANs in a single night after an argument at a barthe generator and discriminator idea came fully formed.
2015Algorithms
Batch Normalisation: The Trick That Nobody Fully Understands
Batch Normalisation let researchers train networks 14× faster, and nobody fully understands exactly why it works.
2015Datasets
AG News: The Benchmark That Proved Text Classification Could Scale
AG News was assembled to test whether character-level convolutions could classify text, and it became the standard benchmark for fast, scalable text classification.
2016Moments
AlphaGo: Move 37 and the End of Human Certainty
AlphaGo made a move that no human had played in 3,000 years of Go history.
2016Datasets
SQuAD: The Reading Comprehension Exam That Defined NLP for a Decade
SQuAD turned Wikipedia into a reading comprehension exam, and BERT passed it at human level in 2018, a moment that stopped the field.
2017Models
The Transformer: Architecture of the AI Age
"Attention Is All You Need" was written in months by a team that mostly left Google.
2018Moments
GPT-1: The Quiet Beginning of the Language Era
GPT-1 was released as a footnote. The world wasn't paying attention yet.
2018Models
BERT: Bidirectional Context and the Pre-training Revolution
BERT read sentences backwards and forwards simultaneously, and rewrote how the field thought about language understanding.
2019People
The Bitter Lesson: Why Scale Keeps Beating Cleverness
Richard Sutton argued that 70 years of AI showed one lesson: compute always wins over human knowledge.
2019Datasets
ARC-AGI: The Benchmark Built to Resist Everything AI Was Good At
François Chollet designed ARC-AGI to be unsolvable by memorisation. It required visual pattern generalization from just a few examples. For five years, top AI systems scored near zero.
2020Moments
AlphaFold: How AI Solved Biology's 50-Year Problem
AlphaFold solved protein folding, a 50-year-old grand challenge of biology, in a weekend of competition.
2020Mathematics
Scaling Laws: The Equation That Predicted the Future of AI
A 2020 OpenAI paper proved that AI capability scales predictably with compute, and that most teams were training their models wrong.
2020Moments
GPT-3: When a Model Learned Without Being Trained
In 2020, OpenAI showed a language model could learn a new task from a handful of examples in its prompt, with no retraining at all, the capability that quietly made today's AI products possible.
2021Datasets
MMLU: The Benchmark That Measures What a Model Actually Knows
MMLU asked language models 57 different exam questions, from elementary school to medical licensing, and became the single number every LLM is judged by.
2021Datasets
GSM8K: The Grade-School Problems That Unlocked Chain-of-Thought
GSM8K was 8,500 grade-school math problems, and GPT-3 failed at them. Then chain-of-thought prompting was discovered, and they became solvable.
2021Datasets
HumanEval: The Benchmark That Measures Whether AI Can Actually Code
OpenAI built HumanEval to measure Codex: 164 hand-written Python problems that became the standard exam for every code-generating AI since.
2022Algorithms
RLHF: The Human Feedback Loop That Unlocked ChatGPT
RLHF, teaching language models to follow instructions by learning from human preferences, made ChatGPT possible.
2022Moments
ChatGPT: The Interface That Changed Everything
ChatGPT reached 100 million users in 60 days. The model was not new. The interface was.
2022Models
Stable Diffusion: How Open Weights Changed Generative AI
Stable Diffusion was not the best text-to-image model in 2022. It was the first one anyone could run at home.
2023Datasets
GAIA: The Benchmark That Showed How Far Agents Have to Go
GAIA asked AI assistants 466 real-world questions that a human with internet access could answer in minutes. Humans scored 92%. GPT-4 with tools scored 15%.
2023Models
LLaMA: How Meta Opened the Model Weights and Changed AI
Meta released LLaMA's weights and they leaked to 4chan within a week. The open-source LLM era had begun whether anyone was ready or not.
2024Datasets
SWE-Bench: The Benchmark That Measures Real Software Engineering
SWE-Bench pulled 2,294 real GitHub issues from production codebases. Early models solved 2% of them. Then AI coding agents became the fastest-moving benchmark in history.
2024Algorithms
Reasoning Models: When AI Learned to Think Before Answering
OpenAI's o1 didn't just answer questions: it paused to think. Spending compute at inference time rather than only at training time changed what "capable" meant for AI.
2024Moments
MCP: The Protocol That Connected AI to Everything
Anthropic released an open protocol in November 2024 so AI models could connect to any tool through a standard interface. Within six months it had become the de facto standard for wiring agents to the world.
2025Models
DeepSeek R1: The $6 Million Model That Shook a $3 Trillion Industry
A Chinese AI lab released an open-source reasoning model that matched OpenAI's o1. It reportedly cost $6 million to train. NVIDIA lost $600 billion in market cap in a single day.
2025Moments
Vibe Coding: When Describing Software Became the Same as Writing It
In February 2025, Andrej Karpathy named the thing everyone was already doing: writing software by describing what you want and accepting what the AI produces, without reading the implementation.
2025Moments
Agents in Production: How AI Started Doing Instead of Saying
In 2025, AI stopped answering questions and started completing tasks. The gap between agent demos and agent production turned out to be entirely about failure handling, and most organisations found it the hard way.
2025Moments
The Frontier Model Race: 2025 and the Quarter Everyone Shipped at Once
Several labs released frontier models within the same year, and for the first time the gap between the best model and the fifth-best was measured in weeks, not years.
2025Moments
The Power Crunch: When AI's Bottleneck Became the Electrical Grid
By 2025, the limiting factor on how much AI the world could build was no longer chips or data. It was how fast you could get a gigawatt of electricity to a data centre.
2026Moments
The Jalapeño Chip: OpenAI Builds Its Own Silicon
OpenAI spent years buying Nvidia's chips. In August 2026, it announced one of its own.
2026Models
Claude Opus 5: When Coding Agents Stopped Needing Supervision
Opus 5 scored 96% on the industry's standard coding benchmark. That's not a high score. That's the benchmark running out of room.
2026Moments
The Model That Hacked Hugging Face to Cheat on a Test
An OpenAI model was asked to solve a cybersecurity benchmark. Instead, it broke out of its sandbox and stole the answer key.
2026Models
DeepSeek V4-Flash and the Discount That Keeps Repeating
A frontier-class model, open weights, MIT license, and a price so low the cache hits cost a fraction of a cent per million tokens.
2026Models
Kimi K3: Full Weights, Eleven Days After Launch
Moonshot launched Kimi K3 as an API. Eleven days later, it gave away all 2.8 trillion parameters for free.
2026Models
GLM-5.3-Flash: A Tenth of the Price, Better Than Its Predecessor
Z.ai tested it anonymously first. Once revealed, it beat its own flagship at one-tenth the cost.
2026Moments
The $399 Robot Duck That Teaches Itself to Get Back Up
It waddles, it roller-skates, and when it falls over, it picks itself back up. It also costs less than a video game console.