GPT 5 release date 🔥 might be closer than we think | OpenAI applies for GPT-5 Trademark in the US.

Key Takeaways

so open AI is supplied for the trademark of gpt5 at the United States patent and trademark office they have to disclose some of the features that will be available in this model let's take a look at what we can expect while we don't know when this will be out and Industry Insider spills some info about why gbt5 might be coming this year yeah this year and why some people are pretty nervous about it so first here's what openai claims will be in GPT 5. it will include software 4 the artificial production of human speech and text conversion of audio data files into text voice and speech recognition machine learning based language and speech processing now gbt4 we believe took somewhere around 60 million to Train That's not including research costs just imagine rows and rows and rows of Nvidia cards as far as the eye can see running 24 hours a day for months that's the part that costs 60 million but we also believe that there were many inefficiencies in the process of training gpt4 and since then the cost of these cards dropped significantly while their speed and power has increased many full so it's important to understand that if openai was just to spend the same amount and effort to reproduce gpt4 today and improving whatever issues caused the inefficiencies of the first run just redoing the training would massively improve the capabilities that gpt4 had by the way we do have a video covering all the leaked gpt4 model info if you're interested I'll leave it in the description below but my point is that this isn't likely to be an incremental Improvement another thing to keep in mind the gpt4 was released March 14 2023 but that's not when it was complete that was when it was finally deemed to be ready for worldwide release this thing could be well over a year at this point so the thing that is rolling out is something that's going to be based on all the things open AI learned all the data they collected from the world using their products for all of this time from all the improvements that Nvidia was able to push out during this time improving the hardware available Etc so the thing that's going to be rolling out is hard to even imagine openai seems to be always way ahead of where anyone can predict but one guy that seems to know a thing or two so this is from Twitter take this with a grain of salt but blader AKA siki Chen has shared some info from what appears to be an inside source now sticky Chan is behind Runway financial and is on the ground in San Fran investing building and hobnobbing with all the AI insiders here's what he said I have been told that GPT 5 is scheduled to complete training this December and that open AI expects it to achieve AGI artificial general intelligence which means we will all hotly debate as to whether it actually achieves AGI which means it will meaning if we can can't all agree it means it's there it continues so how does an llm even GPT 5 get to AGI don't you still need to prompt it don't it need like to do stuff have goals well imagine you something like this give it a go put it in a loop with all the power of Chad GPT plugins and with GPT 5 instead we are so much closer than people think he also adds also to be clear I don't mean to say achieving a GI of GPT 5 is a consensus belief within opening eye but non-zero people believe it will get there meaning there are some people that believe it will get there he also points to an experiment someone done that basically took cha gbt and looped it with itself where one instance of Chad gbt basically prompts itself in a loop to try to build out a business plan or whatever other prompt it's given and at the end they drop a bomb Oh by the way this isn't gbt4 this is this is GPT 3. 0.5 turbo the faster cheaper and Dumber model and this isn't an isolated case multiple research papers use a similar approach to create incredible success with AI agents we covered several of them on this channel but the point is that stringing together multiple gpt4 instances each one doing one layer of work for example when doing planning when doing reasoning one writing code as needed for example that has been showing incredible promise things like Voyager AI the 25 Chad gpts that run their own village things like that are showing autonomous independent abilities to reach goals iterate change plans and continuously move towards one certain objective getting better and better and better with each new step now people have also been successful with improving gbt4 with prompting improvements for example using things like Chain of Thought tree of thoughts Etc sometimes improving reasoning abilities 20x for some specific tasks adding memory to gpt4 has been showing great promise having something that ranks and sorts these memories allowed gbt4 to be better at synthesizing new information and gaining new insights about existing information so just by throwing whatever thoughts it had into a text file and then for example having another instance of gpt4 rank those bits of information based on how important they are condensing them and then running it through the reasoning gpt4 again has been shown to provide new insights this thing gets smarter as it replays its past experiences and kind of reminisces and ruminates on them my point is that what happens when we take this whole body of knowledge that we've gained over the last less than a year that we've had gpt4 available to Tinker with everything that we've learned up to this point and then we use it with GPT 5 as the core and that's just one but multiple ones each one with its own specific set of tasks all working together towards some one defined objective is that going to be AGI is it gonna be here December 2023 I have no idea but I agree with blader great Handle by the way I agree with him that it will be harder and harder for everyone to agree on whether AGI is here or no whatever is coming next is Gonna Be Wild pay attention this is Wes Roth and thank you for watching