During the talk we learned about some very interesting topics about ChatGPT and an evolution that left Gates “speechless,” as well as the applications that both use the most in their daily lives. However, there was also room to talk about the future of OpenAI language models. And all eyes are now on GPT-5 and the improvements it will offer to the user.
During the talk, Sam Altman was able to talk about all the new features and improvements that OpenAI hopes to implement in the future in its language models. Responding to Gates' question about what OpenAI's goals would be in the next two years, Altman highlighted several points that could be key for GPT-5 to keep OpenAI at the forefront of artificial intelligence.
Multimodality, and greater reasoning capacity and reliability. One of Altman's highlights has to do with multimodality. And the CEO of OpenAI highlights that they are focused on making their language models offer various abilities, such as the possibility of conversing with ChatGPT using voice, offering tools for images and also for video, this last point being one of the still unknown aspects of OpenAI.
In addition, Altman assures that one of the most important areas of progress for OpenAI is in reasoning capacity. According to the CEO of OpenAI, GPT-4 can currently reason in a very limited way, so he expects GPT-5 to excel more broadly in this regard.
Reliability is another very important aspect for OpenAI in its language models and they plan to continue improving in this regard. "If you ask GPT-4 a question 10,000 times, probably one of those 10,000 is pretty good, but it won't always know which one," Altman said during his talk with Gates.
Another area in which OpenAI hopes to improve its language models is adaptability and customization, especially to meet user and business demand at an individual level, causing artificial intelligence to work in ways that users themselves want.
Talking to ChatGPT as if it were just another WhatsApp contact is possible: this is how you can receive their "divine" advice. "People want very different things from GPT-4, different styles, and we will make all that possible; also using your own data, like connecting to your calendar, email can improve the AI model to know when to book an appointment "Altman said.
It is expected that in the coming months OpenAI will finally launch GPT-5, a language model that will replace the current version of GPT-4, which is the most advanced model launched by OpenAI to date. The company has big plans for the next version of its language model, so we will have to be attentive to find out what aspects will end up surprising us.