These last few hours have been a revolution in the artificial intelligence sector. And it is that as we expected, OpenAI officially presented GPT-4, the latest version of its complex language model that offers a great advance in its characteristics, obtaining much more accurate performance and with options that have surprised us all.
Users subscribed to ChatGPT Plus can already try this new version, and since its launch we have already seen truly amazing tests, and that we are talking about in just a few hours. Perhaps the most interesting of all was shown to us by the OpenAI team itself during the presentation. And it is that this model of language is able to interpret an image from a simple sketch and turn it into a web page.
GPT-4 and its great ability to interpret images
As we had already anticipated, one of the novelties of GPT-4 is its ability to interpret images, being able to communicate with this artificial intelligence through photographs that we can provide. As they showed us during the presentation, the AI interpreted a simple sketch of the structure of a web page, and by indicating to the artificial intelligence the guidelines to follow, it wrote the necessary code to be able to develop the web page as indicated. in the sketch.
The test website used in question was a simple joke website that clicked on a button to reveal the second part of the joke. Although the structure of the test page was very simple, the fact that the AI could develop the code for it in seconds is already amazing.
As stated in the presentation, interacting with GPT-4 is talking to a neural network trained to predict what the next step to take is. In the proposed situation, he was offered a input partially completed, and the AI determined how to develop it so that the result was as similar as possible to the given guidelines, in this case the sketch of this simple web page. All this thanks to the billions of parameters used to train this AI.
With this idea in mind, anyone with zero programming knowledge can develop very useful projects. The case of Ammaar Reshi, head of design at Brex, is also very representative. And it is that managed to develop a version for browsers of the legendary Snake game with GPT-4 in less than 20 minutes.
Can GPT-4 code an entire game for you? Yes, yes it can.
— Ammaar Reshi (@ammaar) March 14, 2023
Everyone who subscribes to ChatGPT Plus can already try the capabilities of GPT-4, the language model that powers OpenAI’s conversational AI. Developers who want to test this technology and experiment with its API can now sign up for a waiting list.
In Genbeta | The new thing from Microsoft is a ChatGPT capable of “reading” images and generating others on demand: this is VisualGPT and you can try it now