The huge success of ChatGPT, which reached 100 million users in two months and generated excitement for generative artificial intelligence, raised expectations. For several months, digital specialists and the media have been speculating about the launch of GPT-4, the language processing model supposed to succeed GPT 3.5, on which the chatbot ChatGPT is based. After several rumors, this artificial intelligence software was finally unveiled on Tuesday March 14 by OpenAI, the company that has become a specialist in these models capable of generating stunning texts from questions or written requests.
Also listen ChatGPT: gadget or artificial intelligence revolution?
To praise his progress, OpenAI highlights that GPT-4 obtained a score of 90% on a test to become a lawyer in the United States. GPT-3.5, he had not exceeded the mark of 10%, underlines the company. However, the expectations of the most enthusiastic were somewhat dampened: GPT-4 is certainly, as expected, “multimodal”, that is, it is able to receive images – in addition to text – in its requests.
However, this feature is not yet operational in the unveiled version. Moreover, the latter is not, at this stage, accessible to the general public, like ChatGPT, but limited for the time being to a few testers, in a paid version, with a waiting list, and to a few companies who wish to s Use it to develop services, through an interface (API). Finally, GPT-4 does not escape certain flaws of chatbots, such as factual errors, even if they are a little reduced.
“In everyday conversation, the distinction between GPT-3.5 and GPT-4 can be subtle, recognizes OpenAI in its presentation. The difference appears when the complexity of the task reaches a certain threshold. GPT-4 is more reliable, creative, and able to handle more nuanced instructions than GPT-3.5. » Queries submitted to the model by the user can now be up to 25,000 words, which makes it possible to include entire documents, in order to obtain an analysis, a comparison or a summary, for example.
GPT-4 responds better to questions asking for a “reasoning”, also provides OpenAI. Thus, he knows how to find a time slot to stall a meeting between three participants when he is given a paragraph describing their schedule. It would also be more efficient than its competitors in the management of non-English languages, and even rare ones.
Less transparent communication policy
You have 66.16% of this article left to read. The following is for subscribers only.