After the popularity that ChatGPT has achieved in recent months, OpenAI, the company behind this intelligent chatbot, launches GPT-4. This new version of its language model claims to be much more powerful, and even “exceeds ChatGPT in its advanced reasoning capabilities.”
More reliable, creative, collaborative, and better able to handle instructions with many more nuances. This is how OpenAI defines GPT-4an improved version that will exceed the possibilities it offers ChatGPT and that will allow a more natural and human-like interaction.
For this new version, the company has continued to train this Artificial Intelligence with a large amount of data available on the Internet so that it continues to learn and is able to provide more precise information and answers thanks to the algorithms it uses. Compared with the previous version, GPT-4 can produce more detailed responses using up to 25,000 words, up from 4,000 previously.
It should be noted that, for this training, he has used the cloud of azureso it has used this network of Microsoft teams in the cloud, with supercomputers with A100 chips from Nvidia.
In addition, GPT-4 will also be able to generate new content in both text and image formatsalthough to enjoy this option, users must be paid subscribers to the platform.
Microsoft technology at the heart of GPT-4
This alliance with Microsoft has allowed the company to make its search engine Bing already incorporates GPT-4. In the ‘Chat’ section, also accessible from ‘bing.com/chat’, users will be able to benefit from the improvements that the search engine incorporates since it has been using GPT-4 for some time.
This will lead to some of the results it shows including pages found by Bing. In addition, we can also take advantage of the possibilities of this new version through the Bing app for Android, or using Edge as the default browser whenever our computer is Windows.
With this new version, OpenAI seeks to improve the defects that GPT-3 still incurred, on which ChatGPT is based. Despite the many uses in which this chatbot is proving to be very useful, it has been proven that it can sometimes answer user questions with incorrect or even false information.
GPT-4 seeks to solve this problem with much more precise and certain answers. According to the company, it is able to get 40% more score in veracity tests and it has 82% less chance of responding to user requests with content that is not allowed.
Besides being able to enjoy these benefits on Bing, it is also possible to get the most out of GPT-4 through the ChatGPT website, being able to opt for ChatGPT Plus. This is a paid option whereby users can subscribe for $20 per month, which provides priority access to servers and unlimited queries.
What to do with GPT-4
Without a doubt, ChatGPT has revolutionized the panorama given the possibilities it allows in multiple fields. Possibilities that multiply thanks to GPT-4. Asking it to answer questions, generate texts, do translations, or write about specific topics with specific nuances are some of the many actions that users are already testing.
Beyond the controversy generated by the possibility that students can use it to ‘do a job for them’, something to which a solution is being sought with tools that detect when a text is written by an Artificial Intelligence or a humanthere is another no less striking issue.
The possibility it offers to generate code also offers many advantages to developers, and disadvantages, when it is cybercriminals who use it when they want create malicious code.
However, it seems that the benefits that GPT-4, and its future versions, are expected to bring will also lead to the emergence of new professions such as ‘instructional engineer’. These will be in charge of giving orders to these AIs to generate code more quickly and speed up developments.
In addition, with GPT-4, users will enjoy much more complete responses, with more up-to-date data, more detail, and faster. The fact that this new version allows communication with Artificial Intelligence through images will allow more possibilities when it comes to identifying objects or questions related to image-based content.