📖 Glossary

AGI means an AI system that can handle any intellectual task a person can, rather than performing well only on the kinds of work it was built for.

An AI agent is a system that uses a language model to pursue a goal across multiple steps, deciding for itself which tools to call and when it is done.

Algorithmic trading uses computer programs to execute trade orders automatically based on predefined sets of rules and market conditions.

Alignment is the problem of getting an AI system to pursue what we actually want, rather than a proxy for it that happens to be easier to measure.

An Application Programming Interface (API) is a set of rules and protocols that allows different software applications to communicate and exchange data with each other.

A backlink is a hyperlink from one website to a page on another website.

A benchmark is a fixed set of test questions used to compare AI models — useful for tracking progress, and routinely over-read as a measure of usefulness.

A blockchain is a decentralized, distributed digital ledger that records transactions across many computers in a way that makes them difficult to alter.

Cloud computing delivers on-demand computing services, including servers, storage, databases, networking, software, analytics, and intelligence, over the internet.

The context window is how much text a model can hold in view at once — the conversation, the documents, the instructions and its own reply, all counted together.

A core banking system is the software that holds the accounts and the balances — the ledger every other banking service ultimately writes to.

Core Web Vitals are a set of specific metrics from Google that measure user experience aspects of web page loading, interactivity, and visual stability.

A data breach is any incident in which information that was supposed to stay private — customer records, passwords, medical files, source code — is accessed, copied or exposed by someone who was not meant to have it.

DeFi, or Decentralized Finance, refers to a system of financial applications built on blockchain technology that operates without central intermediaries.

Edge computing processes data closer to its source, rather than sending it to a centralized cloud or data center.

An ETF, or exchange-traded fund, is a basket of assets packaged as a single share that trades on a stock exchange, so you can buy a slice of an entire index, sector or commodity in one transaction.

Fine-tuning is the process of taking a pre-trained machine learning model and further training it on a smaller, task-specific dataset to adapt its performance to a new domain or application.

A firewall is a filter that sits between a network or device and the outside world and decides, packet by packet, which traffic is allowed through and which is dropped.

Generative AI refers to artificial intelligence models capable of producing new and original content, such as text, images, audio, or code, that did not exist before.

Generative engine optimization is the practice of getting your content cited inside AI-generated answers, rather than ranked as a link beneath them.

A GPU is a processor built to run thousands of simple calculations at once, which is why it became the hardware artificial intelligence runs on.

Guardrails are the checks placed around an AI system to constrain what it can output or do — sitting outside the model, because the model cannot reliably police itself.

Hallucination occurs when an artificial intelligence model generates outputs that are factually incorrect, nonsensical, or fabricated, despite appearing confident and coherent.

Inference is the process where a trained artificial intelligence model uses new, unseen data to make predictions or generate outputs.

KYC is the regulated process of verifying who a customer is before letting them use a financial service, and monitoring afterwards for activity that does not fit.

A Large Language Model is an artificial intelligence system trained on extensive text datasets to understand, generate, and process human language.

A layer 2 is a separate network that processes transactions off a main blockchain and periodically writes a compressed record back to it, to gain speed and lower fees.

Machine learning allows computer systems to learn from data and improve their performance on tasks without explicit programming.

Malware is any software written to do something the owner of the device would not agree to — stealing data, encrypting files for ransom, or handing control to someone else.

MCP, the Model Context Protocol, is an open standard that lets an AI model connect to outside tools and data — files, databases, browsers, other apps — through one common plug instead of a custom integration for each.

A multimodal model handles more than one kind of input or output — text, images, audio, video — inside a single system rather than by bolting separate tools together.

A neobank is a financial institution that operates exclusively online, without traditional physical branches, offering banking services entirely through digital platforms.

A neural network is a computational model inspired by the structure and function of biological brains, designed to recognize patterns and learn from data.

An NFT, or Non-Fungible Token, is a unique digital identifier recorded on a blockchain that verifies ownership of a specific item or piece of content, which can be digital or physical.

Open Banking is a system that allows customers to securely share their financial data from banks with authorized third-party providers through application programming interfaces (APIs).

Open source refers to software whose source code is made publicly available, allowing anyone to view, modify, and distribute it under specific licensing terms.

Penetration testing is a controlled, authorised attack on a system, carried out by security professionals to find the weaknesses a real attacker would exploit, before the real attacker does.

Phishing is an attack that persuades a person to hand over credentials or run something harmful, by impersonating someone they have reason to trust.

Prompt Engineering is the practice of designing and refining inputs, known as prompts, to guide artificial intelligence models to produce desired outputs.

Proof of work secures a blockchain by requiring participants to spend real computing effort to add a block, making it expensive to rewrite history.

Quantum Computing is a new computing paradigm that uses principles of quantum mechanics, such as superposition and entanglement, to process information.

Ransomware is malware that encrypts a victim's files or locks their systems and then demands payment, usually in cryptocurrency, for the key to get them back.

Retrieval-Augmented Generation is an artificial intelligence technique that enhances large language models by retrieving relevant information from an external knowledge base before generating a response.

ROI measures what an investment returned relative to what it cost, expressed as a percentage — a simple ratio that hides everything the ratio leaves out.

SaaS is software delivered as an ongoing service over the internet, paid for by subscription, with the vendor running the infrastructure rather than the customer.

Schema Markup is a standardized vocabulary of tags and attributes that website owners add to HTML to help search engines better understand the content on their web pages.

Search intent is what someone is actually trying to accomplish when they type a query — and matching it matters more than matching the words.

SEO is the process of improving a website's visibility in unpaid search engine results to attract more relevant traffic.

A smart contract is a self-executing agreement where the terms between parties are directly written into lines of code, stored and executed on a blockchain.

A cryptocurrency designed to maintain a stable value relative to a specific asset, typically a fiat currency like the US dollar.

Staking involves locking up cryptocurrency assets to support the operations and security of a blockchain network, typically in exchange for rewards.

In artificial intelligence, particularly large language models, a token is a fundamental unit of text or data that a model processes.

Tokenization is the process of representing ownership of a real asset — property, a bond, a fund share — as a transferable record on a blockchain.

Training data is the collection of text, images, or other examples a model learns from — and the single biggest determinant of what it can and cannot do.

The transformer is the neural network design behind essentially every modern language model — its key idea being that each word is interpreted in the light of every other word at once.

Zero trust is a security model that verifies every request individually, regardless of whether it comes from inside or outside the network.