GPT-4: Artificial Intelligence that competes with humans in university exams

The vertigo continues.

Oliver Thansan
Oliver Thansan
16 March 2023 Thursday 20:48
6 Reads
GPT-4: Artificial Intelligence that competes with humans in university exams

The vertigo continues. The evolution of GPT-3.5, the AI ​​language model that powered the ChatGPT bot is called GPT-4, and it has quite a bit more power than its predecessor in many tasks. One of its most impressive features is that it is a multi-modal AI: it can work with text, image and sound, although the company OpenAI currently only allows access to text chat to paid subscribers, for which yesterday in vain open use

To see some of its visual capabilities, the OpenAI development team showed in a video yesterday how GPT-4 is able to identify a web page sketch made with a few hand strokes on a paper and write perfectly the HTML code that makes it possible.

GPT-4's skills with images are put to the test. One of the ones being tested is Be my eyes – be my eyes –, an application that allows the blind to receive precise descriptions of anything within reach thanks to the camera on their mobile phone.

GPT-4's texts are deeper than the previous version, and their comprehension, too. A devastating example is that, subjected to university bar tests, the AI ​​obtains a grade within the top 10% of human students, when its predecessor was in the bottom 10%. In all kinds of tests, the machine passes.

Not that she's perfect. It actually has some of the problems that GPT 3.5 had. Not only can it make reasoning errors, but its authors point out that there are also ways to bypass some of the stronger barriers OpenAI has put in place to prevent it from having certain biases.

In fact, they warn that, as was the case with GPT-3.5, it "continues to be completely unreliable", as it is capable of "hallucinating with the facts and making errors of reasoning". "You have to be very careful when using the results of linguistic models, especially in high-risk contexts."

The arrival of the new model has caused a great stir. You can write programming code with great ease. Among the examples that jumped on social networks it was possible to see that it is able to program in a few seconds simple video games, from Pong to Tetris, by asking it only in text form.

Microsoft, which has made OpenAI AI one of its big bets, explained that the new Bing search engine, which has been working for a few weeks, already integrated GPT-4 from the beginning. In the meantime, Google continues to test.

Alberto Romero, technology analyst and author of The algorithmic bridge, a newsletter specialized in artificial intelligence, points out that in GPT-4 "the most interesting thing about the future of AI is that it is multimodal and at the same time it is better in language than the his predecessor". He believes that, although "it's strange that it's so good at separate tasks, it's a necessary intermediate step if we want AI to be a general intelligence, like ours."

The speed with which big tech giants like Google and Microsoft – the latest to sign up has been Facebook – have begun to direct their efforts towards artificial intelligence contributes to a widespread feeling that we are witnessing a paradigm shift .

Emad Mostaque, founder of Stable Diffusion, a popular generative AI that creates images from descriptions, observes that “Google and Microsoft are betting heavily on generative AI as the core of their future. This is not an 'it's still early', but multi-billion dollar companies that are changing their entire strategy and approach. I don't remember a technological and strategic change as fast and as significant as this". The next steps, after search engines, will be office applications.

The size of GPT-4 is larger than that of GPT-3.5, both in the amount of parameters it uses and in the amount of training data; but this time the company has not revealed the figures. Among other characteristics, the size of the model, the hardware used, the training calculation, the construction of the data set and the training method have not been published.

Almost nothing is known about it". This same observation is offered by several experts. It must be remembered that the company has gone from being a non-profit organization to being a company in search of monetizing its services.

The darkest part of the new model of artificial intelligence is its ability to be used in risky activities. OpenAI hired 50 experts in areas such as cybersecurity, biorisk and international security to test the “adversary mode” model. This has only just begun.