Get ready for a monumental event in the AI industry as we introduce you to GPT-5. Its impending arrival is set to shake the foundations of the entire AI landscape. Following the introduction of GPT-4, we noticed a huge improvement over GPT-3.5, often known as ChatGPT. As an illustration, GPT-4 can now process up to 25,000 words, whereas GPT-3.5 was only capable of handling 3,000 words. Additionally, it demonstrated considerable performance gains on exams like the LSAT and the bar exam, among others. ChatGPT-5 is poised to redefine the boundaries of what AI can achieve. Prepare to be informed and amazed by its remarkable capabilities that push the envelope of technological advancements.
One particularly impressive feature is the integration of multi-modality in GPT-4. Upon its launch, it will empower the AI model to process and generate various types of data inputs, including images. In addition to text, GPT-4 also demonstrates enhanced problem-solving abilities, as evident from its success in tackling puzzles and riddles that were beyond the capabilities of GPT-3.5.
Now, you might be skeptical about discussing GPT-5 when GPT-4 has only been available for a little time, and you are justified in feeling that way. However, there are indications that GPT-5 could already be undergoing training and might even be released later this year. A Twitter user, followed by Sam Altman and boasting a notable background in prominent tech companies, recently tweeted that they have been informed about GPT-5’s scheduled completion of training in December, with OpenAI anticipating it to achieve AGI. We won’t delve into the AGI aspect just yet, but let’s delve into what they’re stating here.
This is merely a solitary individual on Twitter expressing this, so take it with a Sam Altman-sized grain of salt. However, what if I inform you that there is indeed some evidence supporting their assertion? What many individuals may not be aware of is that GPT-4 was finalized well before it became public knowledge. In the GPT-4 technical report, OpenAI revealed that they dedicated 8 months to safely research for GPT-4. Consequently, the model itself, GPT-4, was actually finished in August of 2022, even before they unveiled ChatGPT to the world. This signifies that they have had a substantial period of time to train GPT-5 and may already have progressed beyond the training phase and into safety testing.
Naturally, the safety testing phase for GPT-5 will likely be equally, if not more, extensive than what they conducted for GPT-4. However, there are other indications that suggest we may already be at that stage. The rumor suggests that OpenAI is training their models using the new Nvidia H100 GPUs, which can significantly reduce the training time required for all the model’s parameters. These GPUs offer a combination of computational power and high-end memory, as highlighted on Nvidia’s website. The Nvidia H100 Tensor Core GPU, leveraging the Transformer engine, enhances AI training by delivering higher throughput and reducing training time by nine times, from 7 days to just 20 hours, as shown in the chart. This represents a significant leap from their previous setup.
But that’s not all. When we consider the recent revolution that Microsoft anticipates, OpenAI to accelerate the release of models to the public, Sticky’s claim becomes more plausible, doesn’t it? Thus, taking all these factors into account, there is a high likelihood that GPT-5 might arrive much earlier than initially anticipated.
However, setting aside those speculations, let’s consider what improvements we can expect from GPT-5 compared to GPT-4 and what new technologies will emerge upon its release. Naturally, I anticipate GPT-5 to exhibit advancements in areas such as creativity, reliability, and overall intelligence compared to GPT-4. In almost every aspect, there are intriguing advancements that warrant our attention. Notably, GPT-5 is expected to process even larger chunks of text and demonstrate improved memory or attention during conversations. Both of these enhancements would undoubtedly showcase significant progress from GPT-4.
But another interesting ability that I believe will get expanded upon is the ability to reflect and reason. Right now, if you go and play with GPT-4, you can ask it certain questions, and if it gets the question or riddle wrong, you can ask it to reflect on its answer. This allows GPT-4 to go back and figure out where its mistake was and greatly improve its accuracy on the next output. This reflection ability has actually led some to say that GPT-4 is showing the sparks of AGI, and I can only imagine how much crazier things will get on this front with GPT-5.
If you have been paying attention to the news coming out of OpenAI and GPT-4, you would have heard that GPT-4 now supports plugins. These plugins are essentially apps for ChatGPT and include things like Wolfram for solving complex math problems and upgrading GPT-4’s ability to display and organize data. Browsing plugins give GPT-4 the ability to access the current internet and push past its cutoff date of September 2021. Even apps like Expedia allow GPT-4 to help you book and organize a vacation. But these are technologies that we have right now. So, how will GPT-5 expand upon this innovation?
I expect GPT-5 to greatly expand the envelope of plugins and start to push ChatGPT towards the idea of a sort of everything app. What I mean by that is that GPT-5, with its upgraded compute and training parameters, could make the integration of further apps and technologies even more seamless and act as a sort of hub for just about any general task that you need to do on your phone. It could even build towards being something like its own operating system. I expect it to build upon the synergy that we are currently seeing sprout between large language models and general third-party applications.
The reason for this, I believe, is that OpenAI is actually building towards this in hopes that they’ll be able to compete with the likes of Google and even eventually Apple as they build out this ecosystem. But that’s just speculation.
Now, we are about to get serious. A big piece of OpenAI news that recently happened and flew under the radar was the announcement that OpenAI invested roughly $23.5 million into a company called 1X to back the production of their humanoid robot, Neo. 1X, formerly known as Holiday Robotics, is an engineering company that develops and hopes to produce commercially available humanoid robots. They have demonstrated their engineering prowess with the creation of a humanoid robot on wheels known as Eve, which is able to do menial tasks and act as a sort of companion around the house.
The investment from OpenAI, though, is targeting the development of a robot called Neo, which is actually going to be a direct competitor to Tesla’s Optimus robot. And if you look at this thing, it almost looks exactly like Optimus. Now, to me, and I don’t know the details, but this seems like a pretty foolish move as I’m a big believer in the potential of Tesla’s robot for a number of reasons, which you can learn by watching some of my previous videos. But Tesla has a gold mine of data related to real-world machine vision and navigation that I think OpenAI could definitely leverage.
But I digress. So why would OpenAI go after this investment? It’s because I believe they want to build towards AGI, and to do that convincingly, a humanoid frame is going to be critically important. Discussions are already underway regarding whether GPT-4 marks the dawn of AGI. In this context, AGI refers to an artificial intelligence system capable of competently performing nearly any human task and possessing human-level intelligence. However, the truth is that GPT-4 is confined to the domain of a chatbot. If it could manifest in a human-like form, I believe the discussions would heavily lean towards considering GPT-4 as the beginning of AGI.
Supporting this project makes a lot of sense, and I believe a significant hurdle in people’s reluctance to acknowledge our proximity to AGI is the inability to envision it within a humanoid framework. However, if a system surpassing GPT-4 can achieve this feat, our conversations will undoubtedly take on a completely different tone. Are you ready for GPT-5?
