Hey there, AI aficionado! It's your pal Chris here, and boy, do I have a mind-blowing update for you! 🤖

Picture this: Artificial intelligence is revving its engines, ready to transform the job market like never before, and the buzz around town is, "How do we ride this wave?" 🌊

But wait, don't panic! The clever folks at TIME have got your back. They've whipped up a nifty glossary to help you decode the AI lingo and become an expert conversationalist on this hot topic. 📚

And then we took what they did… and made it super simple for you! I would highly recommend bookmarking this one because it’s going to come in handy a lot in the coming days.

So buckle up, and let's dive into the future together! 🚀

The A to Z of Artificial Intelligence

AGI

Researchers tend to disagree on whether Artificial General Intelligence is even possible, but OpenAI and DeepMind are both expressly committed to building AGI.

Alignment

The alignment problem is one of the most profound long-term safety challenges in AI. Some researchers are working on "aligning" AI to human values, but this problem is difficult, unsolved, and not even fully-understood.

Automation

Automation is the historical process of human labor being replaced by machines. The latest generation of AI breakthroughs may result in many more white-collar workers losing their jobs, according to a recent paper from OpenAI and research by Goldman Sachs.

Bias

Machine learning systems can be biased if the data they were trained on reflects social inequities. For example, facial recognition software can work better for white faces than black ones.

Chatbot

AI companies have built consumer-friendly interfaces to allow users to engage with an LLM, or large language model. These interfaces can deceive users into believing that they are conversing with a sentient being, which can lead to emotional distress.

Competitive Pressure

Several of the world's biggest tech companies are racing to launch more powerful AI tools, and some of these systems are already displaying hostility toward users. AI safety researchers worry that this creates a competitive pressure to increase the power of AI systems.

Compute

Computing power is one of three most important ingredients for training a machine learning system, and the more computing power is used to train a large language model, the higher its performance on many different types of test becomes.

Data Labeling

Human annotators describe data before it can be used to train a machine learning system. These workers are often paid barely-above poverty wages and are often required to view and label violent, sexual content, and hate speech.

Diffusion

New tools like Dall-E and Stable Diffusion are based on diffusion algorithms, which learn patterns between pixels in images and their relationships to words used to describe them. These tools can quickly and easily generate photorealistic images.

Emergent capabilities

When an AI is trained on more computing power and data, it can show unexpected abilities or behaviors that were not programmed into it by its creators. These behaviors can be dangerous, especially if they are only discovered after an AI is released into the world.

Explainability

Even the world's most talented computer scientists cannot explain why a given AI system behaves in the way it does, let alone explain how to change it. This is the crux of near-term risks, like AIs discriminating against certain social groups, and longer-term risks like AIs deceiving their programmers.

Foundation model

As AI grows, a divide is emerging between large, powerful, general-purpose AIs, known as Foundation models or base models, and the more specific apps and tools that rely on them. Foundation models are unrestrained and powerful, but also expensive to train.

GPT

The most famous acronym in AI is GPT, which is short for "Generative Pre-trained Transformer". It is a type of neural network algorithm that is especially good at learning relationships between long strings of data.

GPU

GPUs are computer chips that are used for training AI models. The Biden Administration restricted the sale of powerful GPUs to China in late 2022 amid rising anxieties that China might leverage AI against the U.S.

Hallucination

Large language models (LLMs) are trained to repeat patterns in their training data, but the standards for factual accuracy are much lower on web forums like Reddit. This leads to hallucinations, which are causing plenty of headaches for tech companies trying to boost public trust in AI.

Hype

Public discussion of AI is distorted by hype, according to a popular school of thought. This misdirection distracts attention from the real and ongoing harms that AI is already having on marginalized communities, workers, and the information ecosystem.

Intelligence explosion

The intelligence explosion is a hypothetical scenario in which an AI gains power over its own training, rapidly gaining power and intelligence, until humanity goes extinct.

Large language model

When people talk about recent AI advancements, they are usually talking about large language models (LLMs). LLMs are giant AIs trained on huge quantities of human language, sourced mostly from books and the internet, and are capable of many tasks.

Lobbying

Like many other businesses, AI companies hire lobbyists to ensure that any new rules do not adversely impact their business interests. In Washington, the White House has tasked the foundation of Google's former CEO Eric Schmidt with advising on technology policy.

Machine learning

Machine learning is the process of creating AI systems that learn from data. Neural networks are the most influential family of machine learning algorithms.

Moore’s Law

Moore's law states that the number of transistors on a chip doubles approximately every two years, meaning that as time goes on, AI models become more powerful.

Multimodal system

Multimodal systems can receive text and imagery as input and output more than one type of signal. They can act more directly upon the world, but this brings added risks.

Neural Network

Neural networks are the most influential family of machine learning algorithms. They use nodes to perform calculations on numbers that are passed along connective pathways between them, and outputs (predictions or classifications) that increasingly resemble patterns in the original data.

Open sourcing

Open-sourcing computer programs (including AI models) can make it possible for the public to more directly interact with the technology, but it can also lead to additional risks, such as bad actors abusing image-generation tools to target women with sexualized deepfakes.

Paperclips

In some sections of the AI safety community, the paperclip maximizer is an influential thought experiment about the existential risk that AI may pose to humanity. The thought experiment illustrates the surprising difficulty of aligning AI to even a seemingly simple goal, let alone human values.

Quantum computing

Quantum computing is an experimental field of computing that seeks to use quantum physics to increase computing power.

Redistribution

The CEOs of the world's two leading AI labs have each claimed they would like to see the profits arising from artificial general intelligence be redistributed, at least in part. They have not said when or how wide-ranging that redistribution should be.

Red teaming

Red-teaming is a method for stress-testing AI systems before they are publicly deployed.

Regulation

There is no bespoke legislation in the U.S. that addresses the risks posed by artificial intelligence. The European Union is considering a draft AI Act, but it is not legally binding, and there is no significant global jurisdiction currently has rules in place that would force AI companies to meet safety standards.

Reinforcement learning (with human feedback)

Reinforcement learning is a method for optimizing an AI system by rewarding desirable behaviors and penalizing undesirable ones. Human workers or users can rate the outputs of a neural network for qualities like helpfulness, truthfulness, or offensiveness.

Shoggoth

A meme in AI safety circles likens Large language models to shoggoths, incomprehensibly dreadful alien beasts originating from the universe of 20th century horror writer H.P. Lovecraft. The meme takes issue with the technique of Reinforcement learning with human feedback, which results in a friendly surface-level personality but an underlying alien nature.

Stochastic Parrots

The term "stochastic parrots" was coined in a 2020 research paper to criticize large language models for being simply very powerful prediction engines that only attempt to fill in the next word in a sequence based on patterns in their training data.

Supervised learning

Supervised learning is a technique for training AI systems that uses a training dataset of labeled examples to make predictions or classifications. It is useful for building systems like self-driving cars and content moderation classifiers.

Turing Test

Alan Turing devised a test to find out if a computer could convince a human that they were talking to another human, rather than to a machine. Chatbots have become capable of passing the Turing test, but this does not mean they "think" in any way comparable to a human.

Unsupervised learning

Unsupervised learning is one of three main ways that a neural network can be trained, along with supervised learning and reinforcement learning. It is predominantly used to train large language models like GPT-3 and GPT-4, which rely on huge datasets of unlabeled text.

X-risk

AI researchers believe that advanced artificial intelligence may cause human extinction, and that there is a 10% chance that humans will be unable to control future advanced AIs.

Zero shot learning

AI systems often fail to recognize things they haven't seen before, because their training data is limited. Zero-shot learning attempts to fix this problem by working on systems that extrapolate from their training data.

As we wrap up this thrilling adventure, I hope you're feeling pumped and ready to tackle the AI world with newfound gusto! 🎉 Remember, the future is ours to create, and with a little bit of knowledge (and a lot of enthusiasm), there's no limit to what we can achieve together. 🌟

Now go forth and conquer, my fellow AI addicts! May the bytes be ever in your favor. 🖥️

To infinity and beyond! 🚀 

Your digital sidekick,

-Chris Winfield
Founder, Understanding A.I.

P.S. If you found this AI adventure as enthralling as I did, why not help spread the word and share the love? 💌 

Give your friends a chance to embark on this thrilling journey by sharing this free newsletter on social media.

Let's ignite a collective spark of AI curiosity and make this labor of love go viral! 🌐 

After all, who wouldn't want to be part of a world where we're all AI aficionados? 😎