The company behind ChatGPT, an app that generates essays, poems, or computer code on command, released a long-awaited update to its artificial intelligence (AI) technology on Tuesday that it says will be safer and more accurate than before. GPT-4 has been highly anticipated since ChatGPT debuted in late November and impressed users with its capabilities based on an older version of OpenAI technology known as the Large Language Model.

“We created GPT-4, the latest milestone in OpenAI’s efforts to scale deep learning,” the company said, adding that its AI technology “performs at a human level” in some professional and academic tasks.

The company said the model is “more creative and collaborative than ever before” and “solves complex problems with greater precision” than previous models. With the update, text responses from GPT-4 will be more accurate and will come from both image and text input in the future, which is a big step forward for the technology, although this aspect has not been made public yet.

For example, if a user sends a photo of the inside of the refrigerator, the GPT-4 will not only accurately recognize what is there, but also find out what can be cooked with these ingredients. OpenAI said it is working with partner company Be My Eyes to prepare the next step. Much of the new model’s firepower is now generally available through ChatGPT Plus, OpenAI’s paid subscription plan, and an AI-powered version of Microsoft’s currently-tested Bing search engine.

OpenAI is backed by Microsoft, which announced earlier this year that it will invest billions of dollars in the research firm. The Windows manufacturer then quickly integrated the technology into its Bing search engine, Edge browser, and other products. Microsoft’s aggressive adoption of ChatGPT has sparked a race, with Google announcing their own versions of its AI technology, and Amazon, Baidu, and Meta joining to keep up.

Less “hallucination”

OpenAI said that with widely reported interactions with the ChatGPT or Bing chatbot showing users lies, insults or other so-called “hallucinations”, the new version is much less likely to derail than the previous chatbot.

“We spent six months making GPT-4 more secure and consistent. OpenAI said that GPT-4 is 82 percent less likely to respond to requests for prohibited content and 40 percent more likely to respond realistically.

Founder Sam Altman admitted that despite expectations, the GPT-4 “still has flaws, is still limited, and looks more impressive the first time you use it than it does after spending more time with it.”

“Algorithm power will increase, but this is not a second revolution,” said Robert Vezul, CEO of Illuin Technology, a French artificial intelligence startup.

Vezul has questioned the security measures taken by OpenAI, which has faced criticism from billionaire Elon Musk that the company is too reluctant to talk about its AI to avoid further embarrassing backlash.

“I’m not sure if I want an AI to block responses to unknown topics… Should the AI ​​decide whether I smoke or not?” Vesul told AFP.

Other companies involved in the rollout of GPT-4 include Morgan Stanley, which will use artificial intelligence to assist its bankers and customers.

“You actually get instant information from the most knowledgeable person in asset management,” Jeff McMillan said in a statement from Morgan Stanley.

Other partners include online education giant Khan Academy and Stripe, a financial application that will use GPT-4 to combat fraud and other purposes.