You'd think that as the CEO of the company behind ChatGPT, Sam Altman would say good things about its own large language model. It turns out that the executive is not that happy with the capabilities of the AI model, but expects great things from its successor.
Sam Altman on GPT-4
OpenAI is among the frontrunners in the AI race, which means that you can expect great things from its tools and services. There's a reason why the first thing that comes to mind when someone mentions chatbots is ChatGPT.
Still, Altman believes that GPT-4 could stand to get a few improvements. When the executive was asked about the LLM and what it's capable of, he responded by saying that he thinks it "kind of sucks," as mentioned in Gizmodo.
"I think it is our job to live a few years in the future and remember that the tools we have now are going to kind of suck, looking backward at them, and that's how we make sure the future is better," the OpenAI CEO continued.
Instead of building up the hype for GPT-4, Altman talked about its successor instead. He says that the next AI model, which for now we'll call GPT-5, will have the same advancement level as GPT-4 from GPT-3.
The OpenAI CEO said that he didn't want to downplay the accomplishment of GPT-4, but that he did not want to overstate it either. "I think this point that we are on an exponential curve, we'll look back relatively soon at GPT-4 like we look back at GPT-3 now."
Altman being dissatisfied with GPT-4 is not necessarily a bad thing. It could even be a good thing as the tech CEO would work harder to create a more advanced AI model, which would provide more functions in the future.
OpenAI's Innovative Streak
Sam Altman may think that GPT-4 is lacking, but its other projects appear to have turned a few heads. OpenAI's text-to-video generator, in particular, Sora, impressed many people with its ability to create realistic videos.
Using the new AI tool, users can generate videos that are up to a minute long. It's still under development and is being worked on by red teamers. It is being assessed for "critical areas for harms or risks" to make sure that the AI tool would not be misused.
Sora has also been made available to several visual artists, designers, and filmmakers so that OpenAI could gain feedback and learn how to make the model better for creatives. So what exactly is Sora capable of? You'd be surprised with what it can create.
There are still limitations, of course, but even at its early stages, it is already more capable than other AI tools meant for video generating. Other than the subject itself, Sora is able to generate a live and detailed background, even understanding how it could change with the panning of the camera.
In a video demonstration, the camera follows a vintage SUV as it speeds up a dirt road. Upon observing the background, you'll notice that Sora managed to generate trees and other forms of shrubbery in the background, retaining its form as the camera moves past them.
