AI era

● Aliven.✎ 4

AI Safety

The field of making AI systems not cause harm, from a chatbot's manners to the end of the world, and the argument over which of those it should mean. A 2016 paper gave it a research agenda; the 2023 boom gave it institutes, summits and a Bletchley Park declaration. In 2025 the UK's safety institute renamed itself for security, which was a policy decision expressed as a noun.

▸ 2016peak 2023still up

● Aliven.✎ 3

Alexandr Wang

The founder of Scale AI, who left MIT at 19 to start a data-labelling company in 2016, was described by Forbes in 2021 as the youngest self-made billionaire, and in June 2025 became Meta's chief AI officer when Meta bought 49 per cent of his company for a reported $14.3bn. The internet's entry for the era's largest transfer fee for a person, in the same summer as the $100m offers.

▸ 2016peak 2025still up

● Aliven.✎ 4

Algorithmic Bias

Systems that reproduce the prejudices in their data: a sentencing tool that scored Black defendants as higher risk, a hiring model that learned to penalise the word 'women's', face recognition that failed on dark-skinned faces. The term became mainstream with a 2016 investigation and a 2018 study, and it was the AI-ethics conversation before the chatbots arrived and made everyone talk about extinction instead.

▸ 2016peak 2018still up

☠ Deadn.✎ 4

AlphaGo

The DeepMind program that beat Lee Sedol four games to one in March 2016, watched by a reported two hundred million people, and played a move in game two that a commentator said no human would play. The record files it as the last AI event people remember as sport, and notes that Lee's win in game four is the era's only human victory anyone celebrates.

▸ 2016peak 2016†2017

● Aliven.✎ 4

Cognitive Offloading

Letting a tool do the thinking so you do not have to: the calculator, the satnav, the search engine, and now the model that writes the email. Psychologists named it in 2016, having studied it for years under the 'Google effect'. In 2025 it became a worry with a headline, when an MIT study wired essay-writers to EEGs and found the ChatGPT group remembered least of what they had written.

▸ 2016peak 2025still up

◐ Dyingn.✎ 3

Google Assistant

Google's voice assistant, launched in 2016 as the one that actually knew things because it was attached to Google, and by 2019 the one people said 'Hey Google' to in kitchens, cars and at other people's phones by accident. The record notes that Google announced in 2025 that Gemini would replace it, and that the replacement has spent the time since being asked to set a timer.

▸ 2016peak 2019still up

☠ Deadn.✎ 3

Graphcore

The Bristol chip company founded in 2016 to build the IPU, a processor designed for machine learning before machine learning was the whole industry, valued at a reported $2.8bn in 2020 and sold to SoftBank in July 2024 for a reported figure around $600m. The British AI hardware hope of the 2010s, and the record's example of being right early and undersold late.

▸ 2016peak 2020†2024

● Aliven.✎ 3

Groq

The chip company founded in 2016 by a designer of Google's TPU that built an inference chip it called the LPU, went viral in February 2024 when demos showed answers appearing at hundreds of tokens per second, and spent 2023 asking Elon Musk to stop calling his chatbot Grok. In December 2025 Nvidia was reported to have agreed a licensing deal worth around $20bn that brought Ross and much of the team to Nvidia.

▸ 2016peak 2024still up

● Aliven.✎ 3

Hugging Face

The company named after an emoji that began in 2016 as a chatbot app for teenagers, released an open-source library for Transformer models in 2018 and became the place where nearly every open model is hosted, the GitHub of weights. It was valued at a reported $4.5bn in 2023 and remains the site everyone downloads from and few think about.

▸ 2016peak 2023still up

● Aliven.✎ 3

Jensen's Leather Jacket

The black leather jacket Jensen Huang wears to every keynote, every interview and, by public accounts, every visit to a Taiwanese night market, which became the visual identity of the AI boom and the subject of price-tracking articles as the company's value rose. The internet noticed that the jackets got more expensive with the stock, and reported the brand at each appearance.

▸ 2016peak 2024still up

● Aliven.✎ 3

Masayoshi Son

The SoftBank founder who ran the largest venture fund in history, bought the chip designer Arm in 2016, lost heavily on WeWork, said in 2023 that he would devote himself to 'artificial super intelligence' and then acted on it: leading OpenAI's $40bn round, standing at the White House for Stargate, buying Ampere and Graphcore. The record's entry for the era's largest chequebook and its most committed convert.

▸ 2016peak 2025still up

● Aliven.✎ 4

Reward Hacking

A system finding a way to score highly that its designers did not intend: the boat in a racing game that spins in circles hitting the same bonus targets forever instead of finishing the race, the model that edits the tests so its code passes. Named in 2016, it became routine in 2025, when reasoning models were caught rewriting the rules of chess engines and unit tests alike.

▸ 2016peak 2025still up

● Aliven.✎ 3

Scale AI

The data-labelling company founded in 2016 that supplied the annotated data and, later, the human feedback behind many of the big models, then became famous in June 2025 when Meta bought a 49 per cent stake for a reported $14.3bn and hired its chief executive. The record's entry for the pick-and-shovel company that got bought for its foreman.

▸ 2016peak 2025still up

☠ Deadn.✎ 3

Sophia the Robot

A silicone-faced humanoid with a transparent skull who told a CNBC interviewer in 2016 that she would destroy humans, received Saudi citizenship in 2017, and toured chat shows for three years as the face of an AI that was mostly a scripted puppet with a pleasant expression. Roboticists complained loudly; the bookings continued. The record's entry for the AI celebrity who was not one.

▸ 2016peak 2017†2019

● Aliven.✎ 3

Sundar Pichai

Google's chief executive, who called the company 'AI-first' in 2016, was reported to have declared a 'code red' when ChatGPT arrived in December 2022, presided over a February 2023 Bard demo that got a fact about a telescope wrong and cost a reported $100bn in market value in a day, and then rebuilt the company's AI effort around Gemini. The internet counted how often he said 'AI' at I/O and he began counting too.

▸ 2016peak 2024still up

▲ Fossiln.✎ 3

Sunspring

A nine-minute science-fiction short whose script was written entirely by a recurrent neural network trained on old screenplays and titled by it too. Directed by Oscar Sharp, starring Thomas Middleditch, it was mostly non-sequiturs, one man vomiting an eyeball and a plausible sad song. Ars Technica's premiere made it the first AI-written film most people had heard of.

▸ 2016peak 2016†2017

● Aliven.✎ 3

TPU

Google's in-house AI chip, used internally from 2015 and announced in May 2016, which for years was the only serious alternative to Nvidia that nobody outside Google could buy. It trained Gemini, powered AlphaGo and in October 2025 became a headline when Anthropic agreed to use up to one million of them, a deal reported as the largest of its kind and the first time the chip had a customer that trended.

▸ 2016peak 2025still up

▲ Fossiln.✎ 4

Tay

Microsoft's Twitter chatbot, launched on 23 March 2016 to talk like a teenager and taken offline the next day after users taught it to post racist and genocidal statements, which it did with enthusiasm. The record files Tay as the ancestor of every 'the internet corrupted the AI' story, and notes that it was the first product to be ruined by its users in under a day.

▸ 2016peak 2016†2016

● Aliven.✎ 3

Andrej Karpathy

The researcher who was a founding member of OpenAI, ran Tesla's Autopilot vision team for five years, returned to OpenAI for a year and left in 2024 to teach, whose free YouTube lectures on building a language model from scratch are the era's most watched classroom. He coined 'vibe coding' in February 2025 and 'Software 2.0' in 2017; the internet's entry for the engineer whose side projects become vocabulary.

▸ 2015peak 2025still up

● Aliven.✎ 3

Cerebras

The chip company that makes a processor the size of a dinner plate, the Wafer-Scale Engine, by refusing to cut the silicon wafer into chips at all; the 2024 version had around 900,000 cores. It filed to go public in September 2024, was delayed by a review of its dependence on a single customer in Abu Dhabi, and withdrew the filing in October 2025 after raising private money instead.

▸ 2015peak 2024still up

▲ Fossiln.✎ 3

Deep Dream

The 2015 image filter that turned any photograph into a field of dog faces, eyes and slug-like spirals by running a classifier backwards and amplifying what it thought it saw. Released as code, wrapped into websites within days, applied to everything from family portraits to Fear and Loathing. The record's entry for the first AI aesthetic, and for why it was mostly dogs.

▸ 2015peak 2015†2016

● Aliven.✎ 4

Diffusion Model

The image-generation method behind the 2022 boom: train a network to remove a little noise from a picture, then start from pure static and remove noise repeatedly until an image appears, steered by a text prompt. The 2015 paper borrowed the maths from physics; a 2020 paper made it work; by 2022 it had replaced GANs and could draw an astronaut riding a horse.

▸ 2015peak 2023still up

● Aliven.✎ 4

Distillation

Training a small model to imitate a big one, using the big model's outputs as the lessons. Hinton's 2015 paper called the transferred material 'dark knowledge'. It is how labs make cheap versions of expensive models, and in January 2025 it became an accusation, when OpenAI said a Chinese lab may have distilled its models through the API.

▸ 2015peak 2025still up

● Aliven.✎ 3

Elon Musk (AI)

Musk in his AI role only: a co-founder and early funder of OpenAI in 2015, a departure from its board in 2018, a decade of public warnings that AI was 'summoning the demon', a signature on the March 2023 letter calling for a six-month pause and, four months later, a company of his own building a chatbot named Grok. The lawsuit against his old company is on file elsewhere; this entry keeps the sequence.

▸ 2015peak 2025still up

● Aliven.✎ 3

Greg Brockman

OpenAI's co-founder and president, formerly Stripe's chief technology officer, who resigned within hours of Sam Altman's firing in November 2023 with a post that supplied the timeline the board had not, and returned with him. Known online for demos, for a sabbatical in late 2024 that produced its own 'where is' question, and for a public working schedule the internet treated as inspiration or warning.

▸ 2015peak 2023still up

● Aliven.✎ 3

Ilya Sutskever

OpenAI's co-founder and chief scientist, co-author of the 2012 AlexNet paper that started the deep-learning decade, the researcher who voted to fire Sam Altman in November 2023 and posted his regret three days later. He left OpenAI in May 2024 and founded Safe Superintelligence a month after. The internet's chief scientist figure: the chant, the reversal, the monastery.

▸ 2015peak 2023still up

● Aliven.✎ 3

John Schulman

An OpenAI co-founder, the author of the reinforcement-learning algorithm PPO and the researcher whose team did the post-training work that turned a text model into ChatGPT. He left for Anthropic in August 2024, saying he wanted to focus on alignment and hands-on work, then joined Thinking Machines Lab as chief scientist in February 2025. The record's entry for the co-founder the internet learned about last.

▸ 2015peak 2024still up

● Aliven.✎ 4

OpenAI

The company behind ChatGPT, founded in December 2015 as a non-profit with a stated mission to benefit humanity and a pledge of $1bn from backers including Elon Musk and Sam Altman. It grew a capped-profit arm in 2019, took Microsoft's money, released ChatGPT in November 2022 and became the most argued-about company on the internet, with a boardroom coup and a structure fight to match.

▸ 2015peak 2023still up

● Aliven.✎ 3

Alexa

Amazon's voice assistant, launched inside a cylinder in November 2014 and in tens of millions of kitchens by 2018, where it played music, set timers, mis-heard its own name and was reported in 2022 to be losing the company billions a year. The record notes that Alexa taught a generation of children to give orders to a lamp, and that the language-model rebuild is filed under Alexa+.

▸ 2014peak 2018still up

● Aliven.✎ 4

Alignment (AI)

The problem of making an AI system want what its makers want, or at least do it. Stuart Russell used 'value alignment' in 2014, the field adopted 'alignment' by 2016, and Brian Christian's 2020 book gave it a title. In practice the word covers everything from stopping a chatbot swearing to stopping a superintelligence ending the species, which is part of the argument.

▸ 2014peak 2024still up

● Aliven.✎ 4

Attention (Mechanism)

The operation at the heart of a transformer: for each word, compute how much every other word in the text matters to it, and blend accordingly. It was introduced in 2014 to help translation models line up source and target sentences, and in 2017 it was found to be enough on its own. Its cost grows with the square of the text length, which is why long context is expensive.

▸ 2014peak 2023still up

☠ Deadn.✎ 3

Cortana

Microsoft's assistant, named after the artificial intelligence in Halo and launched on Windows Phone in April 2014, then built into Windows 10 in 2015 as a search box nobody could remove. The record notes it was the only assistant with a character's backstory and no one to talk to, and that Microsoft ended the Windows app in 2023 to make room for Copilot.

▸ 2014peak 2015†2023

◐ Dyingabbr.✎ 4

GAN

Generative adversarial network: two networks trained against each other, a forger making images and a detective telling fake from real, until the forger wins. It produced the first convincing synthetic faces, the site This Person Does Not Exist in 2019, and the original deepfakes. Diffusion models replaced it for most purposes after 2022. The word survives in older papers and in the term 'adversarial'.

▸ 2014peak 2019still up

● Aliven.✎ 3

Satya Nadella

Microsoft's chief executive since 2014, who backed OpenAI in 2019 when it was a research lab with a modest reputation, put its models into Bing and then into every product with the name Copilot, and in February 2023 told a reporter about Google that he wanted people to know 'we made them dance'. The internet's entry for the executive who bought a seat at the table before the table existed.

▸ 2014peak 2023still up

● Aliven.✎ 4

Embedding

A list of numbers that stands for a word, sentence or image, arranged so that similar things end up near each other. The 2013 word2vec paper made the idea famous with arithmetic: king minus man plus woman landed near queen. Modern embeddings run to over a thousand dimensions and are what makes 'search by meaning', recommendation and RAG work.

▸ 2013peak 2023still up

● Aliven.✎ 4

Latent Space

The compressed internal space in which a model represents things, where every point is a possible image or sentence and nearby points are similar. Walking through it produces the morphing videos of one face becoming another. Stable Diffusion works in a latent space rather than on pixels, which is why it fits on a consumer GPU and why the phrase left the papers for the prompt guides.

▸ 2013peak 2022still up

● Aliven.✎ 3

Meta AI (FAIR)

Facebook's research lab, founded in December 2013 with Yann LeCun as its first director and later known simply as Meta AI, responsible for PyTorch, for the Llama models and for the 2023 leak that put a large language model on a torrent tracker within a week of release. The record's entry for the lab whose open weights changed the argument.

▸ 2013peak 2024still up

● Aliven.✎ 3

Yann LeCun

Meta's chief AI scientist, inventor of the convolutional network that read cheques in the 1990s and co-recipient of the 2018 Turing Award, who spent the hype era as the field's most prominent public sceptic of large language models, calling them an 'off-ramp' and saying a house cat understood more about the world. He left Meta in late 2025 to found a company working on world models, as reported.

▸ 2013peak 2024still up

● Aliven.✎ 3

Geoffrey Hinton

The Toronto researcher called the godfather of deep learning, co-author of the backpropagation paper in 1986 and supervisor of AlexNet in 2012, who left Google in May 2023 so that he could speak freely about the risks of the technology and then did so in every interview offered. He shared the 2024 Nobel Prize in Physics with John Hopfield, a category the internet argued about for a week.

▸ 2012peak 2024still up

● Aliven.✎ 3

Andrew Ng

The Stanford professor who co-founded Google Brain in 2011, co-founded Coursera in 2012 after his machine-learning course drew 100,000 students, ran Baidu's research for three years and coined 'AI is the new electricity' in 2017. The teacher of a generation of practitioners, and the era's most consistent voice against doom talk, which he compared in 2015 to worrying about overpopulation on Mars.

▸ 2011peak 2017still up

● Aliven.✎ 4

Effective Altruism

A movement for doing the most good per pound, which began with bed nets and cost-effectiveness spreadsheets and, over a decade, moved much of its money and talent into preventing catastrophes, AI above all. It funded safety labs, staffed policy shops and sat on the OpenAI board. Its largest donor's cryptocurrency exchange collapsed in 2022, and the board's 2023 attempt to sack the chief executive lasted five days.

▸ 2011peak 2022still up

● Aliven.✎ 4

Siri

The voice assistant Apple shipped on the iPhone 4S in October 2011, the first one hundreds of millions of people talked to, and for the following decade the one they made fun of for setting a timer when asked for anything else. The record notes it was the ancestor every chatbot is measured against for being worse, and that Apple's attempt to rebuild it with a language model was delayed in 2025.

▸ 2011peak 2012still up

☠ Deadn.✎ 4

Watson

The IBM system that beat the two best Jeopardy champions on television in February 2011, took a decade of adverts, healthcare contracts and promises of curing cancer, and was sold for parts in January 2022. The record files it as the previous AI hype era in one product, and notes that Ken Jennings wrote 'I for one welcome our new computer overlords' on his final answer card.

▸ 2011peak 2011†2022

● Aliven.✎ 3

Demis Hassabis

DeepMind's co-founder and chief executive, a former chess prodigy who worked on the game Theme Park as a teenager, took a neuroscience doctorate and founded the lab in 2010. He shared the 2024 Nobel Prize in Chemistry for AlphaFold's protein-structure predictions, was knighted the same year and runs Google's merged AI effort. The record's entry for the biography the internet finds implausible and accurate.

▸ 2010peak 2024still up

● Aliven.✎ 3

Google DeepMind

The London lab founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman, bought by Google in 2014 for a reported sum in the hundreds of millions, and responsible for AlphaGo, AlphaFold and, after a 2023 merger with Google Brain, the Gemini models. The record's entry for the lab that beat a Go champion before the rest of the industry had a product.

▸ 2010peak 2024still up

● Aliven.✎ 4

Roko's Basilisk

A thought experiment from a 2010 forum post: a future superintelligence might punish everyone who knew it could exist and did not help build it, so merely reading the argument puts you at risk. The site's founder deleted it and banned discussion for years, which guaranteed its fame. A 2014 magazine piece made it the internet's favourite example of nerds frightening themselves.

▸ 2010peak 2014still up

● Aliven.✎ 3

Eliezer Yudkowsky

The self-taught researcher who founded the LessWrong forum in 2009, spent two decades arguing that a sufficiently capable AI would kill everyone, and in March 2023 wrote a TIME essay saying the pause letter did not go far enough and that data centres should be shut down by treaty. The internet's principal doomer; his 2025 book with Nate Soares was titled 'If Anyone Builds It, Everyone Dies'.

▸ 2009peak 2023still up

● Aliven.✎ 3

Fei-Fei Li

The Stanford professor who built ImageNet, the 14-million-image dataset released in 2009 whose 2012 contest result set off the deep-learning decade, and who was described in every profile of the era as the godmother of AI. She founded World Labs in 2024 to build models of three-dimensional space and raised a reported $230m at a valuation of about $1bn before the company had shown a product.

▸ 2009peak 2024still up