Sam Altman Just Revealed OpenAI's Secret Plan

Let's talk about this live stream that OpenAI made some interesting revelations and some optimistic predictions. This was a very informal. The camera is completely out of focus. It's focused on this MacBook here. Completely informal thing which I think would be they want to try and aim to do this every month. They was trying to give insight into where things are going. and they answered some really brutally honest questions like will you just finally admit that chat GBT is a disaster and this and that and the other lots of questions about GBT40 etc etc. So very raw, very uh very honest apparently. But some key things that I picked up on which were quite interesting is that Sam alluded to see this little grayed out thing here. They've alluded to devices. Now we know about the uh pin or whatever it is that that he's working on with Johnny IV. That's a device. This is devices which allude to an idea of multiple things they're looking at. Now we know that chat GBT are throwing everything at the wall and seeing what sticks. They are they are trying to prove that AI is not a bubble and that they can get some money from somewhere. No, no two bones about it. But this was quite interesting for me. It's interesting as well how Atlas is very much a prominent thing here. They've this is Does this make sense to you? This kind of stacking that I was thinking about this. Yeah. But third party applications kind of like taking this the rest of the space doesn't make much sense. So what they're saying here so what third party it's saying third party applications can touch various stages of their platform. Yeah. It can touch data centers, energy and racks. It can touch models. It can open AI account. That's what that means. 100%. Oh, sorry. So generally speaking it does generally speaking it does make sense. Uh CH GP Atlas being an item there is also shows that they are um trying to bet big on it. Another kind of tidbit from this uh stream that I picked up on is that they are predicting AGI by 2028 which no way [snorts] like a it was like a a question from the audience type of thing and it was just you know he spent quite a while thinking about it but uh AGI by 2028. Now, just a reminder, AGI is is completely self-learning AI that can take a brand new and novel challenge and be able to figure it out. Whereas, right now, we're working with an AI that has to learn and it just re regurgitates information that's already learned. Whereas, yeah, AGI and I what's it called? Artificial general intelligence. Yes. is that so very much human like developmental and we're talking about a conscious intelligence. Yeah. And I'm not even um consider maybe the like maybe I'm getting too spiritual but I don't think AGI will ever reach consciousness but I think everyone has their different you know it's a spectrum that people no yes it is a spectrum and we won't like you know like it's not like movies that we have software and then we have a conscious uh intelligence. It's not like from that from this to that overnight. It will be like a gradual learning uh for the systems to get intelligent. But to me AGI has been and always like the definition will be it is intelligent and it is conscious to the point that it is at least similar in some ways to human intelligence. It doesn't have to be, you know, the same tissue and the same logic that we have, but it needs to operate fundamentally on logic. It has to be intelligent in that way and be able to store memory, you know, get basically read and write memory. Uh, and these systems that we have are not that at all. Like we have essentially just a prediction machine. These uh LLMs are just a huge like very very very big and advanced prediction machine and they are just not I I don't I'm not convinced that with the current architecture we can reach AGI. Just the last one here is they're looking to build a 20 gawatt a week factory with each of those gigawatts costing $20 billion. So that's $20 billion a week. Now I know well I don't know actually I've heard that that a lot of these companies are looking to nuclear energy to you know where this energy comes from is is not specul is not said here but this is crazy and they want to get that price down obviously like $20 billion a week is just crazy crazy. A city uses two kilowatts per day. A city. Yes. So this is like insane. So So we are we are talking about like soon AI will uh consume a good percentage of the the total energy that humanity uses. But we are not there by the way yet. And it's we're not even close. It it's using a lot of electricity. Uh but AI is lo using less electricity than most people would probably guess or like media tries to show. Same with water. I've seen that um there are like a lot of people being scared of the amount of water these um infrastructures use but it's not nearly as close as you know to to some like football stadiums or facilities use. Um so yeah yeah um we won't get into it just because of time but like you you've seen Stargate which is a plan to put GPUs and compute in the in space which would be very very cold but uh yeah wild it's very cool it's very cool I like I love the idea like putting those in the space with like big big solar uh panels but we will have like a lot of issues we will have also on earth we have we don't have issue with the solar winds, but up there we have and we need to we're talking about like that level of GPUs up there. And also I I believe that the the lifespan can't be over like um a few years up to a decade maybe. So yeah, there are a lot of problems that need to be solved to make that happen. Soon we will get uh the Dyson sphere by the way. So that's the path to Dyson sphere. This is part of a larger conversation on my show Command AI which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.