OpenAI's next generation large-scale language model, GPT-5, will have better inference capabilities, increased accuracy, and video support, CEO Sam Altman revealed.

On the Bill Gates Unconfuse Me podcast, Altman explained that the next generation model will be fully multimodal, supporting speech, images, code, and video.

During this conversation, he also suggested that the model will address many of the issues related to unreliable responses and models not understanding queries correctly.

"Speech input, speech output. Images. Eventually video," Altman said of future versions of AI models. Obviously, people want that. We presented image and voice, and the response was more than we expected," he explained.

We still don't know much more about GPT-5 than hints from Altman and others. It is expected to be a true multimodal model, similar to Google's new Gemini Ultra.

OpenAI began training GPT-5 last year, and Altman hints that it will be a significant improvement over GPT-4, especially in its ability to understand complex queries and the real world.

Altman told Bill Gates: "It's going to be a steep improvement curve for at least the next five or ten years.

Many large AI labs, including Open AI, have Artificial General Intelligence (AGI) as their ultimate goal. It is to create a superintelligence that is smarter and far more capable than humanity.

There were early indications that GPT-5 could be some kind of superintelligence, but speculation surrounding this model now suggests that it will be a better version of the type of AI we already have, such as GPT-4, Anthropic's Claude 2, and Google's Gemini Ultra It appears to be.

In other words, not only will they have better reasoning abilities and outperform humans in many academic assessments, they will likely have comprehension that more than mirrors human intelligence.

It may also be the next step on the road to AGI. In a speech at Friday's Y-Combinator W24 event, Altman reportedly told the founders and entrepreneurs in the audience that AGI should be built with the mindset that it will happen "relatively quickly."

One of the biggest problems with the current generation of AI models is the fact that they make things up, also known as hallucinations. This is also a reliability problem, which Altman said will be solved with GPT-5.

He told Gates: "If you ask most questions 10,000 times on GPT-4, one out of 10,000 times is probably pretty good, but you don't always know which one is right.

Other important improvements are the ability to customize how the AI reacts, acts, and solves problems. Some of this is made possible by the addition of GPTs (personalized chatbots built on top of ChatGPT).

"People want very different things from GPT-4," says Altman. 'We make it all possible, plus you can use your own data.'

Microsoft and Google have already taken steps to integrate their AI models with personal data through Copilot's integration with 365 and Bard's Workspace.

Altman says this could go even deeper in the future. "The ability to know about you, your e-mail, your calendar, how you want to book appointments, connections to other external data sources, all of that. These will be some of the most important improvements."

I use AI models all the time in my work, playing with different tools, trying to understand how they work and what they can do. giving an AI access to my life, my data, my personality is like asking for trouble, like asking for the emergence of Skynet It's like inviting the emergence of Skynet.