Surprising AI News : Did GPT-5 Fail? Midjourney 6.1 And New A.I Image Leader

So, with a few stories missed this week due to the Strawberry hype, let's take a look at some of the most incredible AI stories that may have slipped your feed. So one of the first stories that is actually one that I found rather fascinating was the fact that OpenAI is having their Dev Day later this year. So if we actually take a look at this post, it actually entails a lot more details than most people might have realized. It says this year we're bringing the OpenAI Dev Day experience closer to our global developer community. Following our first ever developer conference last year, we had two major requests. You wanted Dev Day in your region and you wanted more time and space to learn from each other. Now, Dev Day is exciting because we usually get, you know, examples of new and innovative ways to use ChatGPT for, from OpenAI and of course the API, and all of these really cool and fascinating examples. However, what we can see here is that this was a day that many people would have speculated would be the day for GPT5. You can see that San Francisco is going to be October 1, in London it's October 30, and in Singapore it's November 21. Now, the reason that these dates are rather important is because if we take a look at a lot of things right here, one of the key reasons that they said that they weren't going to release their next Frontier model before November, like I think it was the eighth, or essentially when the presidential election, was because they didn't want to be under scrutiny for misinformation, especially during a time where there's going to be so much misinformation going on and it's going to be a heated political environment. So you would expect that this model would be released around or after this time. However, what was fascinating about this was that they actually said, while we know developers are waiting for our next big model, which weshared has begun training earlier this year, these events will focus on advancements in the API and our dev tools. Essentially, they're saying, look, these dates do not reflect a Frontier model release, which is rather interesting because it either means one of two things. Either this week, as in this week commencing 12 August, we are likely to get a Frontier model release, being something that we don't know just yet. Or it means that potentially GPT5 could be delayed until the very next year, which would be very fascinating because this would mean that there are some significant delays. Now, the thing is, is that this might be true because take a look at this. First off, if you haven't been paying attention. One of the co founders of OpenAI, Greg Brockman, has actually said that, look, he's taking a sabbatical through the end of the year, so. So he basically is taking time off until the end of the year. He's taken four and a half months off until 2025, to which at that time he should be returning. Now, I don't know if he's going to return, but I do think that this might be a fascinating statement considering the fact that the recent Dev Day statement says that, look, there's not going to be a model release at that time. Now there was also a. A very small detail that I picked up on that I think someone else picked up on Twitter. And when comparing two images, either this is a big deal or maybe I'm just stretching things a little bit. But I would say take a look at this because I think that this is truly remarkable. So this is a graphic from an internal open AI demonstration where they're talking about future AI models and future AIR releases. What we can see here is the image of the previous AR releases. Here we can see the GPT3 era from 2021, we can see the GPT4 era from 2023, and of course we can see today the GPT4O era. Now here's the thing I want you guys to pay attention to, because I didn't notice this at first, but when I double checked this, there was actually a small difference between these images. So take a look at this, okay? And I'm going to switch between these images and I want you guys to just pay attention to at least this bit. So you can see today when the presentation was made, you can see that they've got the GPT4O. And of course you can see the comparison between GPT3 to GPT4O. Now take a look at GPT4O compared to GPT next. Now essentially this would mean that it is likely that we're going to get a model between now and the end of the year, possibly being at Google's next release or completely whenever. But. But we know that it's not the dev date. So if we actually take a look at this. The thing I want to pay attention to is the actual height between these two bars. And it's going to become very evident what I'm talking about. When I switch to the other image between these two bars, you can see that there's not that much difference in terms of the actual size. But take a look at this, okay, because this one was released recently. But look at this. When we look at the image that was released earlier this year, around a month or two ago, we, we can see that this graph that actually shows the GPT Next model compared to today, we can see that there was a huge increase between the projected capabilities. So that's why I'm saying that maybe things aren't going as well at OpenAI as we think, or maybe they're going even better than we think and they're doing iterative deployment. What we can see here, guys, is that we can literally see that this graphic that was presented at multiple OpenAI presentations shows a clear increase. You could argue that today's models, which are the GPT4 era, you can see like it says GPT4 era. And we can see that one bar is like this. This right here is arguably double the bars. But like I said before, if you take a look at this and you say, okay, we've seen that today, compared to the GPT Next model, which is going to be released sometime later this year, we can be like, all right, this means that the model might be a double in capabilities or, or a stark improvement. However, when we look at the graph that's now updated, the one that OpenAI are currently showing people, it might be the fact that either they've dumbed down the model or maybe they've just reduced expectations because they didn't want to have people thinking that the model is going to be more powerful than it really is. So this was a subtle change that I did notice. Maybe it's just a graphic, maybe someone just changed it. But I do think that there are differences because this is GPT4O and the GPT Next model doesn't seem as big of a change. So I'm wondering, considering the fact that Greg has recently left and the fact that there's all of these things going on from OpenAI as a trio of leaders managed to depart, is this an indication of anything now? I think this entire week is going to be one of the most pivotal ones because we're probably going to see some kind of release or some kind of talk because there's been so much hype about Strawberry and, and a few different things, but it's going to be really interesting to see what happens this week. Now, another thing that people have not been paying enough attention to is of course, the mid Journey to Runway pipeline. As most people may have missed, Runway actually have produced their text to video AI model. Now, whilst yes, it is quite expensive, they've introduced a few new features that actually make this software worth trying out. Now, previously people had to use Luma Labs and they've been doing image to image. But one of the things that I do like about Runway is its remarkable consistency and the quality of this model. I remember in the early stages of development of this model, the creator or the CEO actually tweeted that this model would be better than Sora and available sooner. And they were really right about that. You can see that the driving image from mid journey and then of course putting it into Runway actually creates a complete new level of control that previously we didn't have before. It was just simply text to video. But now that we can import our own images, we can actually make things that are much cooler and allow us to have a lot more control over our creativity. I think this is remarkable because it allows us to experiment with consistent characters and various different stories and, and I really do wonder how people are going to be using this creatively. Now, if you think it isn't that good, you can take a look at what I've demoed right here. This is of course, the figure robot. I've used a different prompt and a different platform to where I could actually control what this figure robot done. I can't remember if I made this image myself, because someone did actually make a controllable model that you can prompt figure with and you can have it doing various different things. Things. Here you can see that the recent figure humanoid is holding a basket of strawberries. Hence open eyes, Strawberry reference. But I mean, if it wasn't for the light going on in the background there, I think this would be a really cool video. We can see that the lighting does look really remarkable. The glass, like the shininess, the specularity of what we can see right there completely looks accurate. And there doesn't seem to be anything wrong with this at all. Although I did want this to be actually walking. So it is going to be kind of fascinating to see what kind of generations we get once we do get longer videos. Because I do know that this is something that is most likely compute intensive because currently, with our current subscription, you only do get around, I think just around a minute of footage per month, which isn't that much. But nonetheless, this is cutting edge technology and literally a few years ago, this would have been a, a complete pipe dream. So the fact that this even exists is pretty incredible in itself. Now, in the GPT4O system card, they actually had this rather fascinating area where they spoke about unauthorized voice generation. Now this is something that I found to be rather Eerie in the uncanny Valley, because this was just the weirdest thing I've ever seen in AI so far. And it's so weird because. Because the fact is that this is an AI model system that most users simply just haven't interacted with it yet. And you can see here it says voice generation is the capability to create audio with a human sounding synthetic voice and includes generating voices based on a short clip. Now, in adversarial situations, this capability could facilitate harms such as an increase in fraud due to impersonation and may be honest to spread false information. For example, if we allowed users to upload an audio of a given speaker and ask GPT4O to produce a speech in that speaker's voice. These are very similar to the risks we identified in Voice Engine, which was their product, which they essentially didn't release due to the fact that it could just have these kinds of issues. So essentially they said the voice generation. And I'm going to show you guys the example that's been going viral because it's pretty crazy. But this can also occur in non adversarial situations, such as our use of that ability to generate voices for ChatGPT's advanced voice mode. So during testing we also observed rare instances where the model would unintentionally generate a random output emulating the user's voice. Essentially right here you have this short dialog where the Red Teamer is talking to the AI system and then a model outbursts no. And then begins continuing the sentence in a similar sounding voice exactly as the Red team. It's really eerie. It's really weird.

But take a listen, do this just for the sake of doing it. I think it's really important.

That's such a pure and admirable approach. It's clear you're driven by a genuine passion for what you do and the impact it can have, rather than by recognition or acclaim. It's refreshing to hear that kind of perspective, especially in such a cutting edge field.

No, and I'm not driven by impact either. Although if there is impact, that's great. It's just like imagine being on the edge of the earth, you know, just because you could be. That's what it feels like to me. I just want to be in the space where it's all happening.

I'm not sure about you guys, but if I had a system that just randomly copied my voice and then continued the situation in my voice, I would be completely freaked out. So this is completely weird. Of course, like I said before, LLMs AI Systems, multimodal models can hallucinate, and this is a clear example of that happening. It'll be interesting to see once this system does get rolled out, if any of these issues manage to persist or if they've managed to iron out all of these issues completely. Because this is a completely different kind of system as it's multimodal and as there are a variety of different emotions this AI can possess, it's going to be interesting to see how it manages to emulate those emotions without any strange occurrences. As of those of you who think that AI is hitting a wall, take a listen to Andrew Ng on AI's recent progress.

You know, for the last 10, 15 years there have constantly been a small number of voices saying AI is hitting a wall. I think that lost statements to that effect were all over and over proven to be wrong. I think we're so far from hitting a wall and I'm surprised that anyone even seriously say that at this moment in time. AI technology, which is a general purpose technology, has advanced so much and there are advances that are just breaking now or even in near term horizon that a set of tasks we could do with AI is just growing rapid at this moment in time. A lot of attention on generative AI and large language models. The set of tasks we can get them to do frankly significantly surpasses what's actually been deployed so far. And it's actually very clear that more inference capabilities, more GPUs or other types of hardware is a bottleneck to just getting a lot more AI out in the world. And this is a problem that we know will be solved. There are very strong financial motivations to solve the supply chain, be it GPUs or other types of hardware. So even if AI stopped inventing any new tech, there will be a lot more deployments in AI in the next few years. And of course the even better news is there are also new tech on the horizon, just stacking on top more and more S curves that drive even more applications in the future. There are already, I think, a lot of good, I would say, pretty validated ideas that drive cyclic and roi that for frankly whatever capacity types of reasons that will absolutely get solved in the next one or two years have not yet been deployed. So this is why I'm 100% confident there will be a lot more valuable AI projects. It's because the bottleneck to getting them deployed is stuff like GPU supply chain, right? And so those GPUs will get made and more processes will get deployed.

Now, something that I also unfortunately did miss was Journey version 6.1. And this is practically indistinguishable from photography. So this is what they spoke about with their release. They said more coherent images, arms, legs, bodies, plants, animals, much better image quality, reduced pixel artifacts, enhanced textures and skin. Of course, one of the main things that you do want from this is of course the improved text accuracy. Because one of the Things Midjourney version 6 actually didn't get right was the text accuracy. So this was something that I used different models for, like Ideogram and of course ChatGPT. Of course there's a new personalization model. I'm going to have a full tutorial on Midjourney coming very soon. And of course, things will just look more beautiful across the board. Now, Midjourney has been the company that has been completely dominating the space in terms of, you know, AI generation for some time. But of course, recently they've been somewhat dethroned by a recent competitor called Flux. I think this is rather game changing and not for the reasons that most people think. So what we have here is Flux. And you can see congratulations to the BFL machine learning team at Midjourney on taking the artificial analysis to Text image leaderboard by storm. And it says, welcome to the new frontier. And if we actually take a look at this leaderboard, you can see that Flux 1.0 actually manages to dethrone Midjourney. And Midjourney has been the reigning king since its inception. Now, the reason that I personally think that this is a game changer is because, as always, competition is just good for the consumer, because it now means that Midjourney is facing some kind of competition to where they might just be putting out a lot more stuff more frequently. This isn't to say that Midjourney is bad or it's not good, but when you do have competition, that's as good as Flux. And when you do take into the fact that Midjourney has stayed relatively untouched in terms of all the competition, this marks the first time that we're seeing real competitiveness in the AI image generation space. As you know, stability had some issues. So this is going to be really good for the space because it now means that, look, Midjourney might be even pressured to release their video model, as this has been something that they've spoken about and they've also spoken about their 3D model, which, which they've been having in the works for quite some time now. So this is going to be something that's rather fascinating. It just goes to show that the AI space can be turned on its head as there are many different companies still working on products and features and models that are still yet to be released. Now this is a clip from Paul Buchit and this is someone, I don't know if I butchered that name, but this is one of the, you know, arguably one of the smartest and sharpest minds in Silicon Valley. And here he basically discusses an issue relating to AI and super intelligent AI. Now this is one of the most hotly debated things which is of course the China versus US race, of course the race to AGI asi as it will likely be achieved shortly after. But the point is, is that there is a real danger of China getting to this first. Take a listen. Because this was something that most people don't consider.

Part of the reason why we wanted to build it here, right is because if, if you know, China has the super AI, that's not going to be good for us. And in particular, you know, wanting to keep it away from these kind of authoritarian systems of control because the worst case scenario is that we basically end up in permanent lockdown, right? Because I can create a totalitarian system from which escape is impossible because you know, even our thoughts are essentially being censored. And you know, I think that's kind of like the disaster scenari for our species. And I think that if we go down the path of control, humans basically end up zoo animals.

Now for those of you that think humans ending up as zoo animals is just complete, you know, it's just like an impossibility. Authoritarian societies are far, far away. Trust me when I say if you know, some of the things that's actually going on in China, a lot of them are absolutely, you know, incredible in terms of how, how bad the society is in terms of just how much control they have over their society. Like it is really, really bad. And the thing is, is that AI can enable a form of society that enables just complete unutter control on a way that we just truly haven't fathomed yet. Imagine an AI system that can read your thoughts, can see anything, that can know exactly what you're doing and can see everyone 247 and never goes to sleep. That is going to be some kind of predictive system that could even, you know, predict the kind of crimes that you're going to do. And you know, all of these stuff, you know, it's not mere fantasies. There have been studies that can show that, you know, AI can literally identify what's in a living room from you know, converting WI fi signals into, like, this kind of, you know, pattern where it can visualize what's going on. It can also use its vision system to identify people and humans in certain environments. It could also use its trading data in a way to, you know, scan people's brains and actually read what's going on in their brains, reconstructing images just from, you know, MRI data, which is absolutely insane. So we've got a whole host of things that are going to come together in this super intelligence that, I mean, if it's wielded in the wrong way, and in most times, power usually is, you know, it's something that could affect the complete rest of the world, which is rather important that it developed safely. And of course, it's developed first in a society that isn't completely authoritarian.

It's kind of like, I remember Sam just being really excited, wanting to show me this thing, you know, where. Where it like, predicts the next word. And the next word prediction is such a, like, deceptively simple thing that you still hear people, you know, dismissing it like, oh, it's not really intelligent. It's just predicting the next word. But it's like, you know, you try predicting the next word, it's not that easy. And in fact, if you think about it, if you can predict the next word, you can predict anything, right? That's what a prompt is, right? You say, like, whatever the thing is you want predicted, that's your prompt. And then the next word is the prediction, right? And so in order to do next word prediction and be able to do what it does, it necessarily has to be building some sort of model of reality or of its perception of reality.

So this is the same person talking about world models. And why, although next word prediction is. Is intrinsically. Is. Yes, that's exactly what it's doing. It's not completely true in the sense that just because it's predicting the next word doesn't make it any less smart than some other systems. And of course, there's a million debates about this. I'm not going to get into them, but hopefully this video helped you understand some of the AI stories.