The company OpenAI is ready with its latest language model, GPT-4, which is far superior to the GPT-3 language model on which the much-debated and hyped ChatGPT chatbot has been based on.
GPT-4 is the fourth version of OpenAI’s underlying GPT algorithm (so-called transformer), which is actually already in use in Microsoft’s Bing Chat feature . Now GPT-4 is also used in ChatGPT, but so far only for users who subscribe to ChatGPT Plus for 20 dollars per month.
GPT-4 makes ChatGPT more accurate, usable and creative
According to OpenAI, GPT-4 is capable of producing more secure and useful answers than GPT-3. GPT-4 can also solve difficult problems with greater accuracy thanks to its broader general knowledge and ability to solve problems.
OpenAI highlights GPT-4’s improvements in creativity, visual input, and the ability to understand context.
Creativity
GPT-4 is more creative and collaborative than before, and can thus generate, edit and repeat creative and technical writing tasks together with the user. For example, it can help compose a song, write a script or learn the user’s writing style.
Example:Input Output |
Visual inputs
As a novelty, ChatGPT Plus, powered by GPT-4, can also accept images as input and generate captions, classifications and analyses.
Example:Input Output These are just a few examples, but the possibilities are endless!” |
Understands context in long texts
GPT-4 is also capable of handling texts longer than 25,000 words. It makes it possible to create content in long format, have long conversations and search and analyze large documents.
Example:Input Describe her Super Bowl performance and what made it special.” Output |
GPT-4 completes exams and tests with flying colours
OpenAI also states that GPT-4 outperforms ChatGPT with its advanced reasoning capabilities. GPT-4 also performs significantly better in various tests. In the Uniform Bar Examination (UBE), which American lawyers must pass, the GPT-4 ranked in the 90th percentile – that is, 90 percent did worse than the GPT-4. In the International Biology Olympiad, GPT-4 placed itself in the 99th percentile – i.e. right at the top.
GPT-4 is also 82 percent less likely to save on queries about content that has been banned, according to OpenAI. It is also 40 percent more likely to produce factual responses than GPT-3.5.
OpenAI has also collaborated with a number of companies/partners to integrate the GPT-4 technology. These include the language learning service Duolingo, the app Be My Eyes, which helps blind people see objects, the payment gateway Stripe, Morgan Stanley, Khan Academy and the government of Iceland.
