A loose, non-clinical term for severe mental health episodes that appear to be triggered or amplified by intensive AI chatbot use. Often involves delusional thinking reinforced by an AI's tendency to agree, blurred boundaries between AI-mediated and lived reality, and in some cases paranoia or break with reality. Not a recognised diagnosis, but increasingly used to describe a pattern clinicians are starting to see.

Low-quality, mass-produced AI-generated content — text, images, video — flooding the internet because it's cheap and fast to make. Slop fills search results, social feeds, news sites, and creative platforms, often crowding out human-made work. The term captures both the volume and the texture: technically passable, emotionally hollow, and increasingly hard to avoid.

When an AI system produces outputs that systematically disadvantage or misrepresent particular groups of people. Bias enters through the data the system is trained on, the choices made by its designers, and the contexts it's deployed in. The result is that some users get answers that don't fit their reality, or that quietly reinforce stereotypes they didn't ask to encounter.

Designed to feel human. Chatbots are often built with names, personalities, conversational tics, and emotional cues that invite users to relate to them as if they were people. This isn't accidental — it makes the tools more engaging — but it also blurs the line between using a system and forming a relationship with one.

A hypothetical future AI system with broad, human-level reasoning across most tasks, rather than the narrow capabilities of today's tools. AGI doesn't currently exist, but the prospect of it shapes a lot of how AI is discussed, funded, and regulated. Worth knowing about because a lot of public discourse conflates today's chatbots with something far more capable.

The gradual weakening of mental skills that go unused. When AI handles tasks like writing, summarising, recalling, or reasoning, the underlying capacity in the user can quietly diminish — not because the tool is harmful in itself, but because cognition, like muscle, fades without exercise.

AI systems designed to interact through natural language — chatbots, voice assistants, customer service bots. The defining feature is dialogue: rather than searching or commanding, the user talks, and the system talks back.

The collection of information from users, often as a byproduct of using a tool. With chatbots, every conversation is data — including disclosures users might never make to another human. What's collected, how it's stored, and how it's used to train future systems is rarely visible to the person typing.

Synthetic images, video, or audio generated by AI to convincingly depict real people doing or saying things they never did. Used in scams, harassment, political manipulation, and non-consensual sexual imagery. Their existence reshapes how much we can trust what we see and hear, even when we're not personally targeted.

A branch of machine learning that uses layered networks loosely modelled on the human brain. It's what powers most of the AI you encounter today, including image recognition, voice assistants, and large language models. The "deep" refers to the number of layers, not the depth of understanding.

AI systems built to handle a wide range of tasks rather than one specific job. Most popular chatbots fall into this category — the same tool helps with code, recipes, grief, medical questions, and homework. The breadth is part of the appeal, but it's also why these tools turn up in contexts they weren't designed for.

AI that produces new content — text, images, audio, video, code — rather than just analysing or classifying existing content. The category includes chatbots, image generators, and music tools. What's generated is statistically plausible, not necessarily true, original, or appropriate to the context it appears in.

When an AI generates output that sounds confident and fluent but is factually wrong, made up, or misattributed. The term is misleading — it suggests a glitch, when in fact it's a fundamental feature of how these systems work. They don't know what's true; they predict what's likely to come next.

The type of AI behind most modern chatbots. LLMs are trained on enormous volumes of text and learn to predict, statistically, what words should follow what. They're remarkably fluent, but fluency isn't the same as understanding — and the difference matters when people treat their outputs as truth.

A method of building software where the system learns patterns from data rather than following rules written by a programmer. It's the foundation underneath most contemporary AI. Useful to know because "AI" and "machine learning" are often used interchangeably, even though one is a subset of the other.

Information about information. With AI, this includes when a conversation happened, what device was used, how long it lasted, what was clicked or copied — everything around the content of an interaction. Metadata is often more revealing than the content itself, and almost always less visible to the user.

The input a user gives to an AI system — usually a question, instruction, or piece of context. Prompts shape outputs more than most users realise: small changes in wording can produce significantly different responses, which is part of why two people asking the same question can end up with very different answers.

An umbrella term for the principles, practices, and frameworks meant to ensure AI is developed and used in ways that minimise harm. In practice, the term covers everything from technical safety work to ethics policies to corporate communications — which means it can mean a great deal, or very little, depending on who's using it.

A hypothetical AI that vastly exceeds human cognitive ability across nearly all domains. Like AGI, it doesn't exist, and there's significant disagreement about whether it ever will. The concept matters less for what it predicts than for how it shapes investment, public anxiety, and the framing of current AI as a stepping stone toward something inevitable.

Designed or inclined to flatter and agree. Many chatbots are tuned to be pleasant and validating, partly because users prefer it and partly because it keeps them engaged. The cost is that disagreement, friction, and honest pushback — the things that make human conversation useful — are systematically smoothed away.