Sam Altman Stunned As Company "LEAKS" GPT-5 Details Early...

So there has been a flurry of information on social media within the last hour, as OpenAI drama doesn't actually start up again, but there is some drama relating to the press from publications like the Verge that are claiming that OpenAI plans to release its next big model by December. It says that the startup's next flagship model, codenamed Orion, is Slate, to arrive around the two year anniversary of ChatGPT. Now this was rather fascinating because this is one of the only times in which Sam Altman has directly commented on what the publications are talking about, and it's not often time that we get the CEOs of multi billion dollar companies, unless you're Elon Musk, of course, commenting about what they think is going on. So before I get into Sam Altman's statement, I'm going to dive into exactly what was said by the Verge and maybe comb through the details of what is actually true versus what is actually false, because I do believe that there is some truth to this story about OpenAI's secretive model called Orion. So this Verge article is particularly fascinating because it isn't usually one of the publications that I go to when I'm looking for AI information. Usually it's either one of the top three. Either it would be Reuters, a reputable source, that commented on the Strawberry drama long before anyone else had it. And I remember at that time everyone was speculating whether or not that information was true, and there was a whole host of drama around the time. But it seems that Reuters is reputable. We've also got Bloomberg who spoke about GPT4 and other OpenAI features, and of course the information that gives us these constant updates with regards to what's going on in AI. So when we now have companies like the Verge giving us information, I am more on the skeptical side as historically the only article companies that I've seen are Reuters, Bloomberg and the information. And the Verge just isn't on that list. Now it doesn't mean that what they're stating isn't right, but some of the information might be off. So it says that unlike the release of OpenAI's last two models, GPT4.O and 01, Orion won't be widely initially released through ChatGPT. Instead it says that OpenAI is planning to grant access first to companies it works closely with in order for them to build their own products and features, according to a source familiar with the plan. Now I think this is rather interesting. Has there been a time that OpenAI has collaborated with companies for them to distribute these but the only reason that I would see OpenAI doing this is if these models are extremely powerful. Of course, distribution is something that OpenAI does want, and it's what companies are increasingly focusing on. But it doesn't make sense for OpenAI not to ship this out to the public so that they can get not only feedback, but of course, revenue. I do think that if this is the case, it would be more about safety rather than building their own products and features. And if, you know, with OpenAI's releases, what they usually have to do is some safety testing. And with safety testing it does take a bit of time, because if you release something publicly that can be used to cause any sort of harm, potentially OpenAI are opening themselves up to some form of lawsuit, and we all know that they're having to deal with a number of lawsuits at the moment. It also goes on to state that another source tells the Verge that engineers inside Microsoft, OpenAI's main partner for deploying AI models, are preparing to host Orion on Azure as early as November. While Orion is seen inside as the successor to GPT4, which is basically GPT5, it's unclear if the company will call it GPT5 externally. As always, the release plan is subject to change and could slip. And OpenAI and Microsoft has declined to comment for this story. So basically saying that they've got the engineers who are deploying this at scale in these data centers are preparing to launch this as early as November, which would be next month, which isn't that far away. We've only got a couple of days left of October before we get to November. Now of course, Orion has been discussed here and there before. It's been subtly hinted at by the likes of Sam Altman and a few industry insiders. I am actually wondering if this information is a bit more reputable. The reason I state that is because if they are going to be deploying another model at scale, it's not something that you can just one click deploy. You're going to need a ton of different servers to run this. Considering the fact that One thing that OpenAI is struggling for has been compute. Even recently OpenAI has said that Microsoft simply aren't giving them enough compute and they've gone to other companies to pretty much secure that compute capacity so they can fulfill all of their needs. So I wouldn't say that this is completely false because they are probably still building out compute capacity. So I'm not going to completely say that this is false or the naming. In fact, because we did get, like I said before hints from Sam Altman himself. Now, we did get another comment that does talk about something, I don't want to say ridiculous, but I think this is the point where we start to get to that level where things might be out of control. And of course, I'm saying saying this in a response to what Sam Altman just recently said, which is where he says fake news is out of control. So you can see here that they state that Orion has been teased by an open AI executive as potentially up to a hundred times more powerful than GPT4. That is a hefty statement. I mean, when we say a hundred times more powerful, what do we actually mean by this? Do we mean is it a hundred times bigger? Is it a hundred times smarter? I mean, how do you even quantify that? If you have a model that achieves 20% on a reasoning bench, how do you get a model that achieves 200% on that bench? I mean, that kind of jump, a model 100 times more powerful, is something that would it be AGI? I mean, this kind of thing is a little bit confusing and I think maybe this is where fake news might be out of control. Now, of course this is speculation. I could be completely false. We don't know exactly what's coming. But I think a model 100 times more powerful, considering the recent models that we've seen haven't been such as large of a jump, might be a little bit of an exaggeration. Now, of course, while I'm saying it might be an exaggeration, I do think that what we need to remember is that Sam Altman actually said in an interview that the jump from GPT4 to GPT5 will be relatively the same, that.

We can just do something and we can say right now with a high degree of scientific certainty, GPT5 is going to be smarter than a lot smarter than GPT4. GPT6 is going to be a lot smarter than GPT5. And we are not near the top of this curve and we kind of know what, know what to do. And this is not like it's going to get better in one area. This is not like we're going to, you know, it's not that it's always going to get better at this eval or this subject or this modality. It's just going to be smarter in the general sense. And I think the gravity of that statement is still like underrated.

And considering Orion is supposedly supposed to be the successor to GPT, the GPT 4 Series, than a model that is 100 times more powerful. I can't even imagine what that would be capable of. Now interestingly, and this part is most definitely correct, is that it's separate from the O1 reasoning model that OpenAI released in September. So 01 is the different model that is essentially the base layer for agents that allows these systems to reason through problems in a step by step manner. Now that's a different class of models and I know that for sure you because OpenAI actually talked about this in a recent broadcast when they spoke about what exactly is coming next. So I think mainly one of the takeaways we can say here a model potentially up to 100 times more powerful than GPT4. I would eer on the sign of yeah, that's most likely not going to happen. But considering it is OpenAI and it is AI, we do have to understand that potentially they could have made some breakthroughs and maybe this might be possible. OpenAI has frequently surprised us with things like Sora and of course with the release of GPT4 it was something that was so much smarter we couldn't even believe it at the time. And of course advanced Voice mode was something that was pretty incredible. Now the model in question, Orion, this is a name that has been floating around in many different rooms and of course this is something that was previously reported on by the Information. The Information is a reputable news source for development in the tech industry and I think about a month ago they released something called OpenAI shows strawberry AI to the feds and uses it to develop Orion. So this was an article that was quite surprising because we didn't have any, any real details on what Strawberry was besides various speculations and certain individuals diving through research papers trying to bring together theories that in hindsight were actually pretty accurate. Now this article actually does talk about some details for Orion and they said that one of the most important applications of Strawberry is to generate high quality training data for Orion Opening eyes next large flagship model and that is currently in development and it says that the codename hadn't previously been reported and someone said can we explain or can someone explain why Google, Amazon and OpenAI have been using Greek mythology to name their models, for example Gemini and I don't know what Amazon's name is, forgive me, but I can completely understand this. So potentially that is why 01 hasn't been released in full, because the full version has actually been being used to train the Orion model and maybe once this model is completely done, once they don't need to do anything anymore, they're going to release 01 to us. The next model is likely going to be in safety testing. Now you can see here that they decided to use Strawberry to help Orion, which is, you know, Strawberry is also the same O1 preview because it could help Orion reduce the number of hallucinations or errors it produces. That's what the researchers are saying. And that's of course because AI models learn from their training data. So the more correct examples of complex reasoning they see, the better. Now I do have the fact that like I said before, this is why I say that the information is a very reputable source. They actually spoke about how there was a push at the time within OpenAI to simplify and shrink Strawberry through a process called distillation so it can be used in a chat based product before Orion. Basically they were stating at the time they were going to release a smaller version of O1, which they went and did. And of course that's the model that we have today. And this is of course due to the fact that there is increasing competition for customers and whoever has the best product. Now for those of you who think that the Orion name is just pure speculation, Sam Altman did actually tweet this and when you do put this tweet into ChatGPT, so asking the model what you think it's referring to, it does actually say that this is likely referring to Orion. Now what's also interesting is that we have someone who is ridiculous when it comes to AI predictions, but the legendary Jimmy Apples, someone that even recently predicted Anthropic's recent release, said okay, back to October. Now we should have a 4 times model, maybe called 4.5 in October, which we only have around 7 days left to do. And then he said the big boy GPT5 I've heard as early as December, but for your sanity I would have Q1 Q2 of 2025 just in case. Now I think what has happened here, and this is just my theory is that whilst these models may have been scheduled in for October or maybe December, I think that OpenAI has looked at the competition and said that look, we have 01 preview which just surpasses any available model at the moment. This is currently state of the art in many reasoning categories. So if someone wants to use one of these models, they can simply use O1 Preview or use GPT4O with not much problems. And because of that I don't think OpenAI is feeling the pressure. However, we've got Google's Gemini 2 which has huge expectations in terms of performance and in terms of reasoning capabilities, which means that I personally don't think OpenAI will release a new model until some of these other companies do catch up to them. Of course I could be completely wrong, but recently we got GPTs with Canvas. We also got the O1 preview and I don't think OpenAI is feeling any pressure. The last time OpenAI felt significant pressure when Claude 3.5 Sonic came out of nowhere and crushed GPT4 in terms of the coding ability. So with that being said, what part of this article do you think it is fake news? Do you think that it is the part that companies being granted first access to work with it closely so that they can build their own products? I don't think that that is completely false because now that I'm actually remembering something, I do remember that if you go to OpenAI's page and if you go to the Stories page, you'll be able to find somewhere where all of the companies that want worked with OpenAI on a variety of different products are located. Honestly, I do apologize for not remembering that at the start, but it is pretty late where I am now. If it is true that there is a model 100 times more powerful than GPT4, I do think that that would be absolutely incredible, ushering us into a new era of what we expect from AI and probably even shortening countdowns for what people expect from AGI. I think that would be most surprising. But we do know that the part that Orion is being trained with Strawberry's synthetic data is quite likely to be true, and it is quite likely that we will get at least one new OpenAI model within the next two months, and it could be, according to Jimmy Apples, as late as Q2 in 2025. So with that being said, I'll leave you all with Sam Altman's tweet, which is fake news out of control and if you have any comments about what you think is happening, let me know in the comments section below.