Will ChatGPT-5 launch in 2025?

What might we expect from artificial intelligence in 2025? This year, we are expecting OpenAI ChatGPT-5, the next generation of the AI language model that powers ChatGPT. But will it come to pass? And what might it be able to do? Here’s Mike Wooldridge who is an expert on machine learning at the University of Oxford…

Mike - So just to recall the timeline. GPT-2 was the first notable release in the GPT series. That was about 2018, 2019. GPT-3 was the breakthrough. GPT-3 was the first really, really impressive large scale, large language model that met a large audience. And that was the technology basically that goes into ChatGPT and ChatGPT is November 2022. Then we see GPT-4 rolled out fairly quickly after that, and everybody has been waiting for GPT-5. Now, I've had lots of conversations with colleagues and friends recently, possibly fueled by a few glasses of wine in some cases, about where exactly GPT-5 is. And I've heard flatly contradictory, very confident, but flatly contradictory statements from people about where they think it is. So one view is that the technology has flatlined and that GPT-5, we haven't seen it yet because it just isn't that impressive. And the people who believe that, their beliefs are fueled somewhat by Sam Altman,the CEO of OpenAI behind the company, behind this technology. His statements that the race to scale just making things bigger was not delivering the same level of advantages that it did with GT-3. And so that's fueling this idea that actually GPT-5 hasn't been released because it's a disappointment. So at the same time, I've had very confident statements from colleagues that 'no, no, GPT-5 really is going to be mind blowing', and it's just that OpenAI and co are busy frantically testing it to make sure that it's really, really, really fit for the market so that when it's released, it really does dazzle. So I honestly don't know which of those two statements is true, but a lot I think is riding on it because the expectations are so high.

Chris - Do you think that we will see it in 2025? And what do you in fact see in 2025? Is that big correction that we're considering must be on the cards with this level of investment, this sort of scatter gun approach to investing in so many of these technologies, which is going to bring some companies financially to their knees if they don't discover the golden nugget of this is the use case that everyone wants? Is the reset coming in 2025 or something else exciting?

Mike - Well, I think we will see GPT-5, I think it's unimaginable that we wouldn't see GPT-5 next year. What it will do and what the scope of its capabilities will be is anybody's guess, my guess would be right now that its language capabilities will not be hugely advanced. I think it will still hallucinate. I think it will still get things wrong. We will still start to see, you know, a few days after releasing, we'll start to see on social media people saying, look at this stupid thing that GPT-5 has done. That's my guess is that will all be there. But at the same time, I think what it will probably embrace is different modalities. So where I think that one of the big battlegrounds for generative AI is over the next couple of years, is around multimodal AI. And that means not just text, but text and images. And of course in GPT-4, we've already seen text and images come together. But things like sound, music, video, being able to upload a video and actually having an explanation of what's going on in the video, not just a transcript of the voices in the video. We can do that now, not perfectly, but we can do that now. But actually what's happening in the video, this is a video about a man taking a golf shot and falling over, you know, this kind of explanation. So multimodal I think is where it's really going to be. At some point in the next few years, and maybe even next year, we're going to have generative AI that will be able to take a prompt and generate a TikTok length video to order. TikTok is one of the biggest social media platforms out there right now. And when we have that kind of capability, we're in a different world, a new world, a completely new world of entertainment. So multimodal all of those different modalities. Being able to produce music, and again, we've seen prototype systems that can do this to order, you know, 'I would like a mashup of Joy Division and Ed Sheeran,' you know, something like that, and produce a song that's a mashup of Joy Division and Ed Sheeran, or an Ed Sheeran song in the Style of Joy Division, something like that. All of those different modalities coming together and what people are going to do with that is going to be very weird and very, very wacky, but it's going to take us into a completely new era of entertainment and media consumption. So that, I think, multimodal is where the battleground's going to be. The end point of that, I think, is virtual reality. So, one other thing that we saw was Apple's attempt, they were careful not to call it virtual reality. They called it augmented reality. And the Apple Vision Pro headset. Now it seems like that hasn't been a massive hit. It hasn't been a failure, but it hasn't been a massive hit. But Apple, I don't think, is going to give up on it. I think they will continue and somebody is going to get that technology right. Virtual reality is such a compelling idea. Somebody is going to, somebody is going to get it. Right. And when generative AI comes together with virtual reality, again, I think it's, we're going to be in a completely new era of media and entertainment and we're going to see completely new forms of entertainment produced there.