Is ChatGpt 4 coming? Everything we know so far
[ad_1]
The new version is currently only available in the paid version of Chatgpt (Plus). Numerous experts, professors, OpenAi client companies and international press journalists (such as the New York Times) have managed to prove it. OpenAi also today presented Gpt-4 on its site, with several tests demonstrating the big step forward compared to the previous version
What changes?
OpenAi says it will make Gtp-4 available on ChatGpt (which today, in our test, is still on Gpt 3.5) and via API to selected companies (with a waiting list). The main advances seem to be two: version 4 is more precise (makes fewer errors) and also can analyze images, describing them in detail and even then answering related questions. For example, in one trial a user gave him a picture of the inside of the refrigerator, asking him what recipes he can make with the foods he sees. And the bot provided some recipes, corrected, with those ingredients. Version 4 remains imperfect, according to what emerged from the tests. He keeps getting confused and making things up. It also remains limited in time, because it was trained with firm data in August (compared to 2021 of the previous version).
The change with the search engine
A limit that Microsoft has overcome by combining ChatGpt with the Bing search engine. OpenAi and various independent tests have confirmed the increase in accuracy, in giving answers to 14,000 questions on various subjects and taken from university or professional exams. He surpassed that to become lawyers in the US, for example. Note how, according to the OpenAi tests, the accuracy in Italian is 84.1 percent, the highest figure after English (85.5 percent). Anil Gehian associate professor of medicine and cardiologist at the University of North Carolina at Chapel Hill, described to the chatbot the medical history of a patient he had seen the day before, including post-hospital complications.
Where you see the first progress.
The description contained several medical terms that laymen would not understand. Gehi asked how he should treat the patient and the chatbot gave him the right answer. He also impressed the ability to describe images, with great detail and also with many descriptive paragraphs of text. He did it for an image from the Hubble Space Telescope, for example. It was also more accurate than before in other areas, such as tax accounting or planning. He also managed to make a correct summary of a long article of the New York Times, a proof that ChatGpt often fails. Sometimes even version 4 gets it wrong, however, as it turned out in tests and even can still blatantly blaze. That is, he continues to have “hallucinations”, a technical term used in the field of artificial intelligence to define whole plant inventions. For example, when asked for website addresses describing the latest cancer research, it sometimes generated internet addresses that didn’t exist. Gpt-4’s advances are due to the use of a larger training database and further fine-tuning with human reviewers. The impact of these improvements on the practical use of Gpt-4 by users and companies, even in third-party products that use the API, will remain to be seen in the coming months.
[ad_2]
Source link
