NOS News•
-
Nando Kastelein
editor Tech
-
Nando Kastelein
editor Tech
The advanced text generator ChatGPT sparked a hype in November last year. It suddenly became clear how good AI (artificial intelligence) already is. To the surprise of the makers themselves, who had put the chatbot online in between the planned work. That planned work is an improved version that came out this week: GPT-4.
It has been clear for some time that AI is one of the most important developments in the tech sector. For example, since last month, the powerful position of Google’s search engine has suddenly come under discussion due to competition from Microsoft.
ChatGPT has integrated it into the Bing search engine. These two tech companies are also not sitting still and came up with updates this week. It shows that developments in this area are moving at lightning speed and the potential is great.
Differences between 3.5 and 4
There are a few essential differences between GPT-4 and its predecessor version 3.5. First of all, the new version can handle a lot more words, allowing the system to ‘read’ a lot more and produce a lot more text.
The NOS asked GPT-3.5 and GPT-4 both three questions. These were the answers (due to the limited length, a maximum number of words has been given in the question):
-
NOS
Slider 1 example GPT-3.5 vs GPT-4 -
NOS
Slider 2 example GPT-3.5 vs GPT-4 -
NOS
Slider 3 example GPT-3.5 vs GPT-4
The most eye-catching innovation is not yet available, but is sensitive. GPT-4 can tell what’s on an image and then do something with it. Due to concerns about abuse, the general public cannot yet test this, but the potential has already been made clear.
During a more than twenty-minute presentation this week, OpenAI president Greg Brockman showed a sketch of a website from a notebook. He then took a picture of it and gave it to GPT-4 with the assignment: build a website. That also happened, including working buttons.
“To the outside world, this seems like magic: you go from sketch to site,” responds AI researcher Judith van Stegeren. As a PhD candidate, she conducted four years of research into creative AI at the University of Twente.
“Under the hood, it goes from photo to text, which is then linked to a task to program a website. They take small steps one after the other.” In addition, she points out that a lot of research has already been done on image recognition, and that it was successful early on.
-
YouTube/Open AI
The outline of OpenAI -
YouTube/Open AI
From sketch to website
ChatGPT unleashed a search engine battle between Google and Microsoft this year. For the first time, the dominant market position is under discussion. But it doesn’t stop there: the parties have also been evenly matched for years when it comes to office applications. Gmail vs Outlook, Word vs Google Docs.
And just like in February, the companies tumbled over each other again to show what they have to offer users. You can later ask to summarize email conversations or turn them into a briefing or presentation. Or how about a summary of a video meeting that you joined later.
The shadow sides
In addition to potential, this AI development also has a dark side. Several, you could even say. For example, the text generator can be misused to spread fake news.
And how do you deal with a system that will also make mistakes? Perhaps in important company information. Human control remains necessary. As a user, you must always be careful that the system does not write down things that are not correct. It is not for nothing that OpenAI emphasizes that humans and computers must do this together.
Technology ethicist Olya Kudina, assistant professor at TU Delft, is also concerned about the speed at which developments are currently taking place. “People still have to learn how to deal with this technology. This creates a situation where there is an unjustified urgency to integrate this into your own products now, otherwise you will be too late.”
It’s like OpenAI is using society as a big laboratory.
According to her, this is a major socio-technical experiment. “It’s like OpenAI is using society as a big laboratory, where they can see people’s reactions and take them into account.”
But there is no question of taking back gas for the time being. In addition to Microsoft and Google, communication platforms Slack and Discord, payment service Stripe and language app Duolingo are also working on integrating ChatGPT.
.