What is AI, in one sentence?

A large language model (LLM) is a very good next-word predictor trained on most of the internet. When you type a prompt, it generates the most statistically plausible response one token at a time. That's the whole trick — everything else (ChatGPT, Claude, Copilot) is an interface on top of that prediction engine.

What's the difference between ChatGPT, Claude and Gemini?

They're all frontier LLMs built by different companies (OpenAI, Anthropic, Google). Day-to-day they overlap heavily. Rough current split: ChatGPT is best for general reasoning and image generation, Claude is best for long-context writing and code, Gemini is best for anything that touches Google Workspace. Pick one, learn it deeply, then branch out.

A prompt is just the instruction you give the model. A good prompt has four parts: role (who the AI should act as), task (what you want done), context (what it needs to know), and format (how the output should look). 'Write me an essay' is a bad prompt. 'You are a UCL admissions tutor. Rewrite the personal statement below to lead with impact. Keep it under 4,000 characters. Return only the rewritten text.' is a good prompt.

A token is roughly ¾ of a word — the chunks LLMs read and write in. Every model has a context window (how many tokens it can hold at once, e.g. 200,000 for Claude) and pricing per token. You don't need to think about tokens day-to-day, but knowing they exist explains why long documents sometimes get truncated.

Will AI take graduate jobs?

It will absorb parts of them. Repetitive analysis, first-draft writing, meeting notes, basic coding — all shifting to AI-assisted. What that means for you: the graduates who get hired are the ones who direct the AI, not compete with it. Being 'the AI person on the team' isn't a technical role; it's a communication and judgement role.

Do I need a maths or computer science background?

No. Using AI at graduate-scheme level is a communication skill — clear briefs in, useful output out. Building AI tools with no-code platforms (Lovable, n8n, Zapier) doesn't require maths either. Only research and applied ML roles need the underlying maths, and those are a different job track.

AI 101 for non-technical graduates: everything you need in one page

Start here: what AI actually is

When people say "AI" in 2026, they almost always mean a large language model — an LLM. An LLM is trained on most of the readable internet and learns one thing very well: predicting the next word given everything before it. That's the entire trick.

ChatGPT, Claude and Gemini are chat interfaces sitting on top of LLMs. Copilot is an LLM plugged into Microsoft 365. Lovable is an LLM plugged into a code editor. Once you see the pattern, the whole landscape gets simpler.

The seven words that unlock everything

Model

The trained AI itself — e.g. GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro. Bigger and newer usually means smarter and more expensive.

Prompt

The instruction you give the model. The quality of your prompt caps the quality of the output.

Token

Roughly ¾ of a word. Models see the world in tokens and get charged per token.

Context window

How many tokens the model can hold at once. Claude can read a 500-page book; older models can't.

Hallucination

When the model confidently makes something up. Real problem. Always check facts and citations.

Agent

An LLM given the ability to take actions — click buttons, call APIs, edit files. The next frontier.

RAG

Retrieval-Augmented Generation. Feeding the model your own documents so it answers from them, not from training data.

Fine-tuning

Training a base model on your specific data. Rarely needed for graduate use — prompting and RAG cover 95%.

The four-part prompt that always works

Role. "You are a Deloitte graduate-scheme assessor…"
Task. "Rewrite the answer below to lead with impact and quantify outcomes."
Context. Paste the answer, the job spec, and the assessment framework.
Format. "Return the rewrite only, under 250 words, no preamble."

This structure alone puts you ahead of 90% of AI users. Most people type "make this better" and blame the tool for the result.

The one thing to do this week

Pick one recurring task in your life — CV updates, cover letters, essay drafts, application questions — and rebuild it as a prompt template. Save the template. Reuse it. That's the difference between "I've used ChatGPT" and "I use AI systematically." Employers can hear the difference in an interview.

Frequently asked

What is AI, in one sentence?: A large language model (LLM) is a very good next-word predictor trained on most of the internet. When you type a prompt, it generates the most statistically plausible response one token at a time. That's the whole trick — everything else (ChatGPT, Claude, Copilot) is an interface on top of that prediction engine.
What's the difference between ChatGPT, Claude and Gemini?: They're all frontier LLMs built by different companies (OpenAI, Anthropic, Google). Day-to-day they overlap heavily. Rough current split: ChatGPT is best for general reasoning and image generation, Claude is best for long-context writing and code, Gemini is best for anything that touches Google Workspace. Pick one, learn it deeply, then branch out.
What is a prompt?: A prompt is just the instruction you give the model. A good prompt has four parts: role (who the AI should act as), task (what you want done), context (what it needs to know), and format (how the output should look). 'Write me an essay' is a bad prompt. 'You are a UCL admissions tutor. Rewrite the personal statement below to lead with impact. Keep it under 4,000 characters. Return only the rewritten text.' is a good prompt.
What's a token?: A token is roughly ¾ of a word — the chunks LLMs read and write in. Every model has a context window (how many tokens it can hold at once, e.g. 200,000 for Claude) and pricing per token. You don't need to think about tokens day-to-day, but knowing they exist explains why long documents sometimes get truncated.
Will AI take graduate jobs?: It will absorb parts of them. Repetitive analysis, first-draft writing, meeting notes, basic coding — all shifting to AI-assisted. What that means for you: the graduates who get hired are the ones who direct the AI, not compete with it. Being 'the AI person on the team' isn't a technical role; it's a communication and judgement role.
Do I need a maths or computer science background?: No. Using AI at graduate-scheme level is a communication skill — clear briefs in, useful output out. Building AI tools with no-code platforms (Lovable, n8n, Zapier) doesn't require maths either. Only research and applied ML roles need the underlying maths, and those are a different job track.

Next step

From "I get it" to "I built it."

Fundamentals are the easy bit. The hard bit is shipping a tool a recruiter can click. That's what the two-week CareerGlowUp Build does — you leave with a working AI tool tied to your target role.

Join the waiting list