With a live broadcast on YouTube at 21pm Italian time, OpenAI presented its new gem, GPT-4, which is on a decidedly higher level than the previous version, launched just a few months ago. Now the “conversation expert” ChatGPT is not only more powerful, but it can also “see”. And this factor can give rise to an unprecedented acceleration in the development of artificial intelligence.
From GPT-3 to GPT-4 a sea passes
ChatGPT, OpenAI's "little electronic genius" that responds to messages with text and code, has quickly become the fastest growing app in history, with over 100 million users per month.
In parallel with its development, user communities and user guides were born (I also published one: if you are interested you can find it on Amazon) and even a “proto-profession”, that of prompt engineer.
However, despite its success, ChatGPT had a few kinks to work out. He tended to "hallucinate", generating texts that sounded plausible but weren't. It reflected prejudices, sometimes it “pierced” the filters for illicit phrases foreseen by its creators.
The new GPT-4-based version solves most of these problems, and improves (much) performance.
Eyes open to the world
The great novelty of this model is its ability to respond to both textual and visual prompts. Think about the possibilities: identifying the author of a painting, explaining the meaning of a meme, creating captions for photographs... The truth is that the field of possible applications expands so much that a possible list becomes gigantic.
But GPT-4 doesn't stop there: it is also significantly more "intelligent" than its predecessor, surpassing its results in several tests, such as those for legal professions (LSAT), those used for admission to American colleges (SAT), and many others. OpenAI claims that GPT-4 it is 40% more accurate in generating truthful content e 82% less inclined to respond to illicit prompts (goodbye “evil” versions of the chatbot).
All the rest is history
We will wonder a lot about the incredible characteristics that generative artificial intelligences will acquire more and more (and more and more quickly). Thanks to its ability to “see”, GPT-4 will power many applications we use every day.
First of all, as mentioned, the new one Chat GPT (in its paid version), now capable of processing texts of up to 25.000 words: it can summarize, write and rewrite, manage entire books. GPT-4 is also part of the search engine Bing. The Khan Academy is using it to create a virtual tutor for students, while Be My Eyes has developed an AI assistant that can analyze and describe photographs for people with visual impairments.
Keeping up with its evolutions will be increasingly complicated.
GPT-4, future prospects
Obviously the system can still be improved, it still has imperfections, but the progress is notable and very rapid. The company is already exploring how to also integrate audio, video and other inputs into future versions of the model. Their goal is for GPT-4 to become a valuable tool for improving people's lives by powering numerous applications.
Soon this device will also tell us what it thinks of our clothes, or recommend the most suitable haircut. It will fuel surveillance systems to which we present our friends and relatives, so that they immediately open only to them. It will make our cars “supercars” that interact in a personal way with the driver. It will be the "narrative voice" of many blind people, and will help them feel more integrated. And who knows what else.
As mentioned, I cannot show you all the arrival points: at most, send you back to the starting point with the official announcement on the OpenAI blog, or with the video of yesterday's presentation, so you can get an idea for yourself.
For now, however, let's say "welcome" to this extraordinary all-rounder who, for those who haven't understood it yet, has already changed our lives.