A Chat GPT was one of the novelties that made the most noise during the last two months of 2022, but that doesn't mean that its creator got where she wanted to. To launch theAI that talks", a OpenAI, specialized in creating artificial intelligences for various tasks, used the GPT-3, a language model that uses deep learning to produce texts that would be written by humans.
But the year 2023 must be marked by the GPT-4, which should allow an important leap forward for the AI that has been growing day after day. Know what to expect right now.
Greater knowledge for ChatGPT
The AI that talks OpenAI appeared on the internet in mid-November 2022 and, shortly afterwards, it was already being talked about a lot due to being able to write texts in a similar way to what you and I would do. To make it work, the developer used a language model based on human interaction called GPT-3. Currently, this technology is in its 3.5 version and has already been implemented in the AI that creates texts.
Without wasting time, the company is already preparing the next generation of the language method for AIs, which should be launched in 2023, with the launch of Chat GPT in the professional market happening in the same year.
Some companies and professionals have already had access to the new version so that advances can be seen in practice and forecasts indicate that it has around 175 billion parameters. That's about the same amount of parameters as GPT-3, but the new generation would bring greater awareness, with "a significant fraction of the internet as a whole".
The main advance would be the possibility of creating texts with even more precision and fluency, as well as we can also expect the artificial intelligence to understand commands in formal and informal language (the same goes for the creation of text).
Gary Marcus, psychologist, scientist and artificial intelligence enthusiast who had contact with people who tested the GPT-4, cited that the new language model is “monstrous and should 'blow' minds” with its now more advanced processing power. It is worth remembering that in its public test, the Chat GPT has knowledge of facts and events up to 2021.
New generation but same problems
The psychologist mentions that, despite presenting important advances in relation to GPT-3, the new generation must still “suffer” from the problem of artificial intelligence having a slightly disconnected view of the real world. It is true that yes, the Chat GPT it can be well used in several tasks, but its creator has already mentioned in an interview that it should not be used for more complex and serious activities.
With this in mind, Gary Marcus cites that it is likely that the new language model released in 2023 will have a lifecycle similar to its predecessor. Many uses can be created, but after a while, it will be necessary to be careful.
While the GPT-4 definitely looks smarter than its previous generations, its internal architecture remains problematic. I suspect what we'll see is a familiar pattern: lots of buzz on its release, followed by closer scientific inspection, and then the arrival of recognition that many problems remain.Gary Marcus, psychologist, scientist and artificial intelligence enthusiast
Another discussion of the use of the idea of OpenAI in practice it would be if, one day, she could clash with Google Assistant, Siri and Alexa. THE Chat GPT it's still giving its first steps and, at least for now, it still doesn't connect to the internet, much less is it as complete as its possible competitors.
But the Google is looking at how it will be used in the future and even issued a “red alert” to its teams so that ways to debunk AI writing are developed. The big fear is that if it gets a significant update, people will stop using the Google Assistant, since the Chat GPT it does not have ads and yet it manages to be free.
Do you think the GPT-4 can make people use just the idea of OpenAI day to day? Tell us us Comment!
See also other features
With information: The Decoder