As previously announced, OpenAI has released GPT-4. The new language model should be a milestone in deep learning. Let’s get to the details, i.e. what exactly GPT-4 can do.
GPT-4 as Advocate and Best Talker
When we talk to another person, they can react to what we say to them, but also, for example, to a photo that we show them. GPT-3 has no such capability and only responds to text messages. GPT-4 is better at this – as an input accepts both text and images. Different types of data can be interleaved. The new language model in the OpenAI description is a multimodal solution that, while less capable than a human in many real-world scenarios, can perform similarly to humans in professional and academic testing.
OpenAI describes the possibilities of GPT-4 with an example legal review. The new linguistic model aims to obtain results close to the 10% of people who take the exam with the best results. For comparison, GPT-3.5 could only boast results at the level of about 10% of the worst passers. GPT-4 has been modified for six months and its developers say it is their best. At the same time, OpenAI points out that the results obtained are still far from perfect.
According to OpenAI, in casual conversation, the capabilities of GPT-3.5 and GPT-4 may not be very visible. The differences are to be revealed when the task received by the artificial intelligence will be well composed. The new language model is intended to be more creative and robust, and to support more detailed instructions.
GPT-4 – concrete examples of the possibilities of the new model
OpenAI’s new language model finds application in programming. Can generate code to create a simple arcade game.
Generate a professional look writings in response to very basic information? GPT-4 shows it can handle it.
manuscript sketch as input and… website design as output. Here are the results of the work of GPT-4.
Telling another person what a meme is about can be frustrating. It follows that even it stops being so funny. However, you can ask the AI for help.
GPT-4 better than its predecessors… but it can still make mistakes
OpenAI points out that despite its capabilities, GPT-4 has similar limitations to previous models. He is not completely reliable and may make errors in reasoning. These are the issues we know about ChatGPT. Artificial intelligence can drive for a while fairly natural conversationto invent complete nonsense or, under the influence of a suggestion, to change your mind about your own abilities. Therefore, you should be careful while using the output data and further check its correctness if you want to use it in any serious task.
source: Open AI, photo: Depositphotos/chetroni
