🌱 Débutant

GPT

Generative Pre-trained Transformer. Language model architecture developed by OpenAI, trained on huge amounts of text to generate natural language.

🎯 Why it matters

Understanding what a TPM is can decode the operation of the majority of current writing and assistance tools. It is the technological architecture that triggered the revolution of the general public generation AI by making the machines able to predict the following word with bluffing fluidity.

💡 Concrete example

GPT-4 is the model behind ChatGPT Plus. GPT-3.5 is available free of charge.

⚙️ In practice

A GPT model is based on the Transformer architecture, which uses attention mechanisms to analyze relationships between all words in a sentence simultaneously, rather than reading them in order. During the pre-training phase, the model learns to guess the following word on massive volumes of text. In practice, when you use it, it generates text in a self-regressive way, that is, it reinjects each word generated in its calculation to produce the following word.

Common mistake: To believe that GPT refers only to the ChatGPT application, whereas it is a family of models and a technical architecture used by many industry players.

❓ Frequently asked questions

It is a network architecture of neurons invented in 2017 that allows AI to process textual data in parallel and to understand the overall context of a sentence in a very effective way.

No, they don't think and have no conscience. They calculate statistical probabilities to determine which words are most logical after your instructions.

🛠️ AI tools to practice with