OpenAI has released GPT-4, the latest large language AI model version. GPT-4 has demonstrated “human-level performance” in many professional tasks, making it an exciting development for the tech industry.
GPT-4 is much larger than previous versions, having been trained on more data and using more information. As a result, it is more expensive to operate, but it is now capable of working with text and images. GPT-4 is already in Bing’s AI bot, and other chatbots will likely incorporate it in the coming weeks. The new model will be initially available to premium ChatGPT members and part of an API.
GPT-4 can accept input in text and pictures; its output is in text format. The most significant innovation of GPT-4 is that it can understand what is shown in an image and consider this data when solving a problem. GPT-4 can even explain the data presented as a diagram. However, image uploading is still undergoing closed testing.
OpenAI claims that GPT-4 will give fewer factually wrong answers, go crazy and talk about taboo topics less often, and even perform better than humans on many standardized tests. But, OpenAI warns that the new software is far from perfect and falls behind humans in many situations. The model still has significant issues in many scenarios. The model still has serious problems with “hallucinations” (making up facts), so it is unreliable in representing facts. GPT-4 still tends to insist on being right when it’s wrong.
Many AI researchers believe that most of the latest advances in AI involve running ever-larger models trained on thousands of computer systems. Such training can cost tens of millions of dollars. Microsoft has invested billions in OpenAI and used the Microsoft Azure cloud infrastructure to train the GPT-4 model. Due to a “competitive environment,” OpenAI did not disclose the model’s particular size or the technology used to train it.
To access the API, you must sign up for a waiting list. OpenAI will charge approximately 3 cents for approximately 750 words of clues and 6 cents for approximately 750 words of the answer. GPT-4 is an example of a scale-up approach for better results, and most of the latest advances in AI involve running ever larger multimodal models trained on thousands of computer systems.
