Lesson 7 of 28 · Language models and chat assistants
What is ChatGPT and how does it work?
ChatGPT is an AI chat assistant made by OpenAI and launched on 30 November 2022. You type a question or instruction, and it replies in natural language. Behind the chat window, a large language model generates each answer one token at a time, based on patterns learned from huge amounts of text and tuning with human feedback.
What the name means
"Chat" describes how you use it: as a conversation, in which you can ask follow-up questions and refine the answer. "GPT" stands for generative pre-trained transformer:
- Generative: it creates new text rather than picking from stored answers.
- Pre-trained: it first learned from a very large collection of text before being tuned for conversation.
- Transformer: the neural-network design, introduced in 2017, that the model is built on.
How ChatGPT works
When you send a message, the app combines it with the conversation so far and with hidden instructions from OpenAI, known as a system prompt, and passes everything to a large language model as tokens. The model predicts a reply one token at a time, which is why the answer appears word by word.
OpenAI's launch announcement explains that ChatGPT was tuned with reinforcement learning from human feedback (RLHF): people wrote and ranked example answers, and the model was trained towards the responses they preferred. Around the model, the app adds features such as file uploads, web search, image creation and running code; which ones you get depends on your plan and changes over time.
What ChatGPT is good at
ChatGPT is a general-purpose assistant. Typical uses include:
- Drafting and rewriting emails, posts, reports and summaries
- Explaining a concept at whatever level you ask for
- Brainstorming ideas, outlines and names
- Translating and adjusting tone between languages
- Writing, explaining and debugging code
What ChatGPT is not
It is not a source of guaranteed facts. It can hallucinate, inventing plausible details, quotes or references, so check anything important against the original source. Without web search, it also knows nothing that happened after its training cutoff.
It is not the only chat assistant, and it is not the same thing as the model inside it. ChatGPT is a product; OpenAI's GPT models also power many other apps through an API. Alternatives include Claude from Anthropic, Gemini from Google and DeepSeek. Each differs in strengths, limits and data terms, so compare them for your own task.
Example
A good first prompt to try
Try this promptI'm new to ChatGPT. Ask me three short questions about my work, then suggest two everyday tasks where you could save me time. Keep each suggestion under 40 words.
This prompt practises two habits that improve almost any answer: giving the assistant context about you, and setting a format and length. It works in other chat assistants too, which makes it a simple way to compare them.
Key points
- ChatGPT is OpenAI's chat assistant, launched on 30 November 2022.
- GPT stands for generative pre-trained transformer.
- It writes replies token by token with a large language model tuned on human feedback.
- It can be confidently wrong, so verify facts, figures and citations.
- It is one of several chat assistants; compare them before you pick one.
Try it on Mindber
Related terms
Related topics
Frequently asked questions
What does GPT stand for in ChatGPT?
GPT stands for generative pre-trained transformer. Generative means it creates new text, pre-trained means it first learned from a very large collection of text before being tuned for chat, and transformer is the neural-network design, introduced in 2017, that the model is built on.
Is ChatGPT free?
ChatGPT has offered a free tier alongside paid plans that add higher usage limits and extra features. Plans, limits and prices change over time, so check OpenAI's own pricing page or Mindber's ChatGPT profile for the current options before you decide.
Is ChatGPT the same as a search engine?
No. A search engine finds and ranks existing web pages. ChatGPT generates a new answer from patterns in its training, and it may also run a web search and summarise the results. Generated answers can contain errors, so open the cited sources when accuracy matters.
Sources
- Introducing ChatGPT. OpenAI, November 30, 2022
- Training language models to follow instructions with human feedback. Ouyang et al., arXiv, March 2022
- Attention Is All You Need. Vaswani et al., arXiv, June 2017
Last updated