The previous generation of the natural language processing algorithm, GPT-3, once made a splash. Artificial Intelligence generated texts that, with minimal tweaking, could be published as if they were human-made. The next step was the virtual interlocutor ChatGPT, which still amazes even specialists. Against this background came the announcement of the release of a completely new language model, GPT-4, and it promises a real revolution in the industry.
The big news came as if out of the blue from Andreas Braun, CTO of Microsoft Germany. He was speaking at the AI in Focus - Digital Kickoff on Thursday, March 9, according to Heise Online. During the presentation the company's representatives spoke about big language models (LLM) like the GPT-series as a breakthrough technology which is radically changing the market. Among other things, they talked about Microsoft's partnership with OpenAI.
It was in this context that Brown announced "the impending release of GPT-4 next week". However, he didn't give any details about the revolutionary neural network. Only in general terms outlined the most notable evolutionary changes to the algorithm.
First of all, the model will support "literally all languages". It is logical to assume that not all of the more than seven thousand existing languages in the world, but the first few dozen of the most popular ones are certainly implemented. According to Brown, a neural network will be able to set a task in German and get an answer in Italian. It is not a question of on-the-fly translation, but of formulating the text immediately in the desired language environment.
Secondly, GPT-4 is now multimodal, that is, it is able to process not only "plain" text, but also recognize it in audio, images and video. It's not entirely clear from press reports, but the impression is that the model will be able to produce a response in the same form.
Perhaps the most curious element of the GPT-4-related event is the slight hint of a practical application for the model. Brown, again, didn't go into detail, but did reveal that this artificial intelligence is already being tested in various applications for Microsoft Azure customers.
For example, a solution has been created for one company that summarises a phone call. It will save a large call centre about 500 operator hours a day. Building the prototype based on GPT-4 took less than two hours and required the efforts of just one developer.
The use of the language model by business customers is considered in three main areas: processing of voice queries by phone, help with paperwork, and answering customer questions that require recourse to a closed set of knowledge. In fact, it automates a whole class of tasks that hundreds of thousands of customer support centre operators now perform.
Recall that GPT (Generative Pre-trained Transformer) is a family of neural network algorithms for natural language processing developed by OpenAI. Since the end of 2015, the firm