“AI’s most advanced system, producing safer and more useful responses” as they market the newest update of the language model.
As an improvement to the ChatGPT, GPT-4 can also understand photos as well as text. What this means is that it can actually analyze the picture and understand what is in it. OpenAI says that GPT-4 is more creative and actually less likely to invent facts.
User can now ask questions about a picture thanks to model’s “multimodality”. It can accept massive amounts of text and remember it, using a whole book as a prompt. Despite of the grown abilities, it can still “hallucinate” answers if it doesn’t quite know the answer.
When shown a picture of a squirrel with a camera and asked why the picture is funny? GPT-4 answer was “We don’t expect them to use a camera or act like a human” which is obviously very accurate. One could also send GPT-4 a picture of a painting and ask, what style does it represent and who is the artist. Though it can not invent pictures itself.
The whole GPT-family is based on GPT-3.5 AI model published 2020. The new features won’t be added to the free version of ChatGPT just yet.