The successor to GPT-3.5 is a fact. GPT-4 adds an extra touch of creativity to conversations and can also analyze images.

After Microsoft gave a subtle hint last week that the launch of GPT-4 was imminent, OpenAI has now confirmed this. GPT-4 is an improved version of the GPT-3.5 language model on which ChatGPT is built. OpenAI instantly integrates the new model with its popular chat tool, although you must have a ChatGPT Plus account to get started. Anyone with access to Bing Chat has also been introduced to GPT-4, Microsoft announced today.

According to OpenAI, conversations with GPT-4 won’t feel much different, but the new model has improved intellectual abilities. ChatGPT can now show more creativity and at the same time argue better logically. In addition, GPT-4 can handle larger blocks of text; up to 25,000 words. You can use this capacity to quickly summarize long texts.

Picture says more than words

Perhaps the biggest difference between the two models is that GPT-4 can also handle images. The AI ​​model can tell you what’s in a photo, but also suggest a recipe based on a photo with ingredients, for example. Combinations of image and text questions are also no problem for GPT-4.

Cum laude

To demonstrate the intelligence of GPT-4, OpenAI took the AI ​​tool in different directions through some university experiments. From law to biology, GPT-4 was often among the top students in the class. In particular, the mathematical skills have improved significantly. Some do fear that GPT-4 is moving towards “general intelligence”, the point at which the intellectual abilities of artificial intelligence have rivaled those of the human brain.

According to Sam Altman, CEO of OpenAI, there is still a long way to go. He has tried to dampen expectations since the first rumors about GPT-4 and continues to emphasize that GPT-4 is far from bug-free. When testing, you will find that the language model sometimes dares to give strange answers. OpenAI has worked for up to six months to increase factual knowledge about GPT-4 and claims it has improved by 40 percent.

Anyone with a paid subscription can already get started with GPT-4. You have to set this manually and OpenAI currently works with a limit of one hundred questions per four hours. Developers can join a waiting list for access to the API.