OpenAI’s new AI model can provide more detailed answers and even interpret images. After months of anticipation and rumors, OpenAI has officially announced GPT-4, its new AI model, which is expected to be behind its popular products like ChatGPT and the new Bing.
Capable of interpreting images as well Unlike GPT-3.5, which is only capable of processing text, GPT-4 can interpret both text and images (although it can still only respond with text): “GPT-4 is capable of solving more complex problems with higher accuracy, thanks to its extensive knowledge database and problem-solving abilities.” For example, you can feed it a picture of ingredients you have at home and ask the model to find a recipe that suits them.
The company showcased a number of examples of verbal problems that were challenging for the current ChatGPT and how GPT-4 can answer them with much greater accuracy. The company emphasizes that the differences between the two models will not be drastically noticeable to regular users, but will become more prominent when you start challenging the new model with more complex questions, or when you need a more detailed response, as GPT-4 can provide up to 25,000 words in a response (as opposed to ChatGPT’s current limit of 3,000 words).
OpenAI has disclosed a lot about the 6-month process of developing the new model, explaining that they worked with 50 “experts” in the initial stages on everything related to safety and model security.
The company claims that GPT-4 is about 82% safer than its predecessor in terms of responding to unauthorized requests.
You can start experimenting with it quite easily. At this stage, if you want to try GPT-4, you need to register for ChatGPT Plus, the premium service of ChatGPT, which will cost you $20 a month (also available in Israel). If you are developing applications or need access to the API, the company has already opened its waiting list, and prices are $0.03 per 1,000 tokens, which is about 750 words.
But you have another way to use GPT-4, as Microsoft has confirmed that Bing Chat, the chatbot developed with OpenAI, is already running on GPT-4. And remember the part where you can feed images to the model as part of the prompt? The company is currently only testing this feature, and it is not yet ready for public testing.
