- GPT-4 is the fourth iteration of Open AI's Large Language Model (LLM).
- A new key feature is that it is "multi-modal", which means it can understand both text and images. This means understand what is in a picture and express logical ideas about it. Eg. It could see a picture of a tethered helium baloon, and undertand that the baloon would rise if the string was cut.
- It has improved guardrails to prevent things like unwanted content generation. Eg. It will refuse to tell you how to make a bomb.
- It has a special 'System Message' that can be sent along with your query. This system message can tell GPT-4 how to behave when responding to any queries asked of it. E.g. You could set the system message to tell GPT-4 that it's name was PirateGPT and that it must respond to any queries like a pirate talking.