Microsoft, at an event in early February 2023, announced that it would integrate OpenAI’s ChatGPT technology into its search engine Bing. It has now been confirmed that the new ChatGPT-powered Bing will utilize OpenAI’s latest multimodal large language model, GPT-4 (Generative Pre-trained Transformer 4), the fourth model in the GPT series. With OpenAI’s generative text chatbot, users receive comprehensive, conversational replies to their questions through an autoregressive language model based on Natural Language Processing (NLP) known as GPT-3, which is a neural network trained using internet data of more than 175 billion parameters to generate text of any kind. OpenAI, however, has upgraded its technology to GPT-4, which is a hybrid training system trained with around 100 trillion parameters, making it more efficient and accurate.
The latest GPT-4 model, is claimed by the company to be “more creative and collaborative than ever before.” Microsoft has confirmed that Bing will now incorporate OpenAI’s GPT-4 model, enabling users to make more efficient searches.
The new AI-powered Bing uses ChatGPT’s latest technology, GPT-4, however, several modifications have been made to the model to make it suitable for search and public use.
Users who wish to try out the new GPT-4 experience on Bing may do so by signing up for its preview. However, on preview mode, when asked what version of GPT it is currently utilizing, Bing AI answered that “the new Bing AI uses ChatGPT version 3.5, but this variant is more capable, accurate, and capable than the version of ChatGPT that you can use in OpenAI.” It remains to be seen whether this answer will change in the near future.
Ever since its free-for-all public prototype launch on November 30, 2022, ChatGPT and AI have been the biggest talking point in the tech world. OpenAI just announced its latest product, GPT-4, currently being deployed by its ChatGPT chatbot. With this new multimodal large language model, not only are there higher data parameters (100 trillion compared to 175 billion in GPT-3), but also improved input methods, algorithms, parameterization, and alignment.