OpenAI last week announced ChatGPT-4, the new version of its chatbot with artificial intelligence (AI). The new technology can now generate more complete and cohesive answers, with more similarities to a text written by humans. Another new feature is the possibility to analyze images by the chatbot, and not only texts. According to the developers, the successor to ChatGPT 3.5 is still 40% more likely to produce correct answers and is less likely to create biased or offensive content. For now, GPT-4 is only available as a paid version, but it is possible to use it on partner platforms that have the chatbot API.
To use ChatGPT-4 on the OpenAI site, you must subscribe to the ChatGPT Plus plan, which costs $20. Below, see everything that is known so far about the new version of the chatbot.
What are the differences between GPT-4 and the initial ChatGPT?
The new chatbot update, in theory, is more efficient than its predecessor. According to the developers at OpenAI, it can give more complete and text-like responses from humans. In addition, the company claims that there are fewer glitches in GPT-4 compared to its predecessor – which reduces, for example, the amount of content considered inappropriate on the platform.
To achieve this result, OpenAI claims that it spent six months training the tool with so-called “adversarial testing programs” and with ChatGPT itself. Despite the improvement, the system is not perfect; therefore, it is still important to check the information from reliable sources and be aware of possible biased or prejudiced texts.
One of the great new features of GPT-4 is the possibility of inserting images into the chatbot, which previously accepted only text. The feature is available as a preview and allows the user to upload images so that the AI can analyze them and create texts related to their content.
The new version of the chatbot also has increased processing power. GPT-4 now has a limit of 32,768 tokens – which in practice allows it to write up to 64,000 characters. Thus, artificial intelligence can maintain longer interactions, as well as analyze long content and generate larger texts.
Finally, ChatGPT-4 surpasses its predecessor in language understanding, improving on some 26 languages. Of course, its greatest ability to interact remains in English – in the language it has 85.5% accuracy – but in less popular languages, such as Telugu, the results were also encouraging, with 71.4% accuracy.
Who can access GPT-4 now?
The main way to use the new version of the chatbot is through a ChatGPT Plus subscription, which currently costs US$20. This can be done by going to “https://chat.openai.com/” (without quotes) and logging in or creating an account on the platform. Then click on “Upgrade to Plus” and on “Upgrade Plan”. Finally, enter your payment information and tap “Subscribe”.
In the new version of the chatbot, developers will be able to access the API tool to make improvements to their applications. Thus, an indirect way to access the technology is through platforms that have ChatGPT-4 integrated, such as Bing, Duolingo, Khan Academy, and Stripe.
How to use free ChatGPT-4?
One of the main ways to use the new version of ChatGPT for free is through the Bing search engine in the Microsoft Edge browser. The system was one of the first to feature a customized version of the chatbot, which helps make search results more accurate. To use this version, simply go to “https://www.bing.com/new” and click on “chat now”.
Another well-known tool that now relies on artificial intelligence is the language learning application Duolingo. The company announced a paid plan called Duolingo Max, which now allows you to train languages from a chat with a chatbot.