The first demos make clear the advances of the new version of the OpenAI natural language model.
GPT-4 has come to completely revolutionize the field of AI-based natural language models. The great update presented by OpenAI brings with it great improvements in its capabilities of interpret queries and generate responsesand we only had to wait a few hours since its announcement to be able to see what he is capable of this model, which It can already be used through ChatGPT and the new Bing search engine.
The first examples and experiments Those who have tested the capabilities of GPT-4 make very clear the improvements of this model in several aspects, and although there is still a wide margin for improvement, the results are impressive. Let’s go over some of the things that GPT-4 can do, using the indications suitable.
Create a web page from a paper drawing in a matter of seconds
Shortly after the announcement of the introduction of GPT-4, Greg Brockman, President and Co-Founder of OpenAI wanted to demonstrate some of the new capabilities of his model through a transmission broadcast on YouTube.
One of the demonstrations consisted of transform a freehand sketch on paperin a functional website using HTML and CSS code.
“See” and describe images
One of the great innovations introduced in GPT-4 is the image input support. Now, the OpenAI model is not only capable of interpreting text: it can also “see” a photograph, understand its content, and generate a description. During the presentation of the model, it was demonstrated how GPT-4 is capable of determine, even, if an image is “funny” depending on what appears in it.
Find security gaps
Previous versions of the OpenAI language model have already been used to discover possible vulnerabilities on platforms, applications or websites. However, there were some limitations in the model, which meant that it was not as effective when it came to detecting this type of breach.
All the improvements carried out on GPT-4 seem to have ended, to a large extent, with these limitations. It has been shown how, in a matter of seconds, the model is capable of detecting vulnerabilities in a contract based on Ethereum technologyand even offer ways to exploit the vulnerability.
I downloaded a live Ethereum contract at GPT-4.
In an instant, it highlighted a number of security vulnerabilities and pointed out superficial areas where the contract could be exploited. He then verified a specific way he could exploit the contract. pic.twitter.com/its5puakUW
— Conor (@jconorgrogan) March 14, 2023
Create games in less than a minute
Several creators have already taken advantage of the power of GPT-4 to build games from scratchwithout the need to be familiar with programming concepts or video game design.
For example, it has been possible create a variant of the mythical “Pong” that can be run in any browser in less than 60 seconds.
I don’t care that it’s not AGI, GPT-4 is amazing and transformative technology.
I recreated the game of Pong in less than 60 seconds.
It was my first try.
Things will never be the same. #gpt4 pic.twitter.com/8YMUK0UQmd
— Pietro Schirano (@skirano) March 14, 2023
It has also been possible to create a replica of “Snake”the legendary snake game of Nokia mobiles, in less than 20 minutes and without programming knowledge. The game is available to anyone who wants to play it.
discover new medicines
Other users have tried to put GPT-4 to a less than playful use. It has been shown that the model is capable of discover new medicines from the composition of a drug. To achieve this, the model find compounds with similar propertiesmodifies them to ensure they are not proprietary, and does whatever is necessary to buy them from those distributors that offer them in their catalogue.
GPT-4 does drug discovery.
Give it a currently available drug and it can:
– Find compounds with similar properties
– Modify them to make sure they are not proprietary.
– Buy them from a supplier (including even sending an email with a purchase order) pic.twitter.com/sWB8HApfgP
— Dan Shipper (@danshipper) March 14, 2023
Analyze complete documents
Thanks to the increase in the limit of words that GPT-4 can interpret, it is now possible to provide entries of up to 25,000 words. This opens the door to full document analysisincluding those that contain text, photographs, screenshots and even diagrams.
It’s only been a few hours since GPT-4 was made official, and we’ve already seen what some members of the community are capable of with the new multi-modal system. And that, which at the moment is only accessible to ChatGPT Plus subscribers.