A scientist specializing in artificial intelligence talks to Sky News Arabia about the chances of these new models to beat Chat GPT soon, whether other models will come out, and the disadvantages and advantages in this race.

OpenAI's Chat GPT, in which Microsoft is investing strongly, came to light on November 30, 2022, and dominated everything else, until its name became synonymous with any mention of artificial intelligence, continuing over time to gain popularity.

The program also continues to break records, including collecting 100 million users within 3 months, gained from users by TikTok and Instagram.

Gemini International and Jess Emirati

In recent months, two models have emerged to compete with GPT Chat: Google's Gemini and UAE's G42 Group's Guess.

Google's AI scientists believe they can beat ChatGPT and are passionately working at the company to take it to the next level on a project codenamed Gemini.

The company will use the AI model previously used to beat the table game world champion to build on the company's already advanced AI tools, hoping to remove the crown from Chat GPT's head.

In the Arab world, the Artificial Intelligence Center of the UAE group "G42" launched an open-source language model in Arabic, under the name "Jais".

Guess is the world's highest-quality Arab competitor to Chat GPT and others in terms of pre-trained AI generating transformers.

GUESS was developed in collaboration between Inception, the artificial intelligence hub of the UAE's G42 Group, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), the world's first graduate university dedicated to AI research, and Cerebras Systems, according to the Emirates News Agency.

Will this accelerated race continue?

A scientist at Amazon at Seattle, Washington, and one of the founders of artificial intelligence, Hisham Iraqi, answers this question that the large number of new advanced models we are seeing and the acceleration of development in them "is much greater than we expected."

  • But Iraqi does not expect the new and upcoming models to be spared from mistakes in this acceleration, saying:
  • I don't think the new models actually trained on all the data on the Internet written and authored by humans.
  • New content won't be as massive as in previous templates.
  • Accordingly, AI companies have begun to develop the idea that humans respond to questions and conversations in these models, and then the model develops itself during use, learning better from human responses to it.
  • Human responses are usually from people who are not familiar with the subject of the conversation, and therefore sometimes they give false information during conversations, and the model is difficult to distinguish between right and wrong.