GPT-5 is Coming: What We Know So Far & Its Release Date

Author(s): UPDF

Originally published on Towards AI.

GPT-4 is the world’s most widely used multimodal language model that has revolutionized AI. However, tech advancements simply have no limits. And that’s why the tech world is already looking for the next big thing, i.e., GPT-5.

Despite the early reports that OpenAI isn’t training GPT-5, CEO Sam Altman has now confirmed that GPT-5 is in progress. This has raised suspicion about what to expect from GPT-5 and when the big launch is coming up. So, let’s dive deep into GPT-5 and discuss all the information we know so far about it.

What Is GPT-5?

GPT-5 is OpenAI’s upcoming multimodal language model. The company hasn’t openly discussed what its next multimodal language model will offer, but Sam Altman has termed it as “superintelligence”. It will be a prominent improvement over GPT-4, especially in its ability to understand complex questions and the real world.

One thing is evident: GPT-5 will elevate natural language processing capabilities via larger datasets and advanced ML. Since GPT-3 and GPT-4 were built on 175 billion and 1 trillion parameters, GPT-5 will definitely use a larger dataset. This will lead to enhanced capability in creativity, reasoning, and versatility.

What Will GPT-5 Offer?

After going through the available information so far on GPT-5, some of the key features/improvements likely coming with GPT-5 are as follows:

1. Competitive multimodality

Multimodal AI systems are booming, like Google Bard and Microsoft’s Bing Chat. To remain competitive, GPT-5 will likely come with comprehensive multimodality. It means that the model will be able to process and generate text, audio, images, video, and similar other content. It will make the user experience more interactive and empower users to do much more than they imagined.

2. Extensive data training

As mentioned earlier, GPT-5 will be trained with a much bigger data volume compared to GPT-4. In fact, OpenAI has deployed a new web crawler, GPTBot, which is designed to crawl the web and gather content to train its models. According to Altman, the training data will be sourced through publicly accessible internet information and by buying from private companies.

3. Better accuracy

Though significant improvements in accuracy were made in GPT-4 compared to GPT-3.5, there are still further enhancements to be pursued. For instance, GPT-4 has around 70% accuracy for code-related queries, so there is much to improve here.

That’s why GPT-5 will offer better accuracy than GPT-4 and make responses more reliable than ever before.

4. Improved contextual understanding and personalization

Another rumored feature of GPT-5 is its improved contextual understanding. It will be able to adapt to a wider conversational context and improve interactions.

In addition, it will also elevate personalized experience. It will be able to learn your preferences and past interactions and provide responses accordingly.

What OpenAI CEO Sam Altman Has to Say About GPT-5?

Sam Altman has been giving hints about GPT-5 for the past year. However, he didn’t provide any technical details mainly because it’s very early for him to validate some of the main features coming with GPT-5.

Recently, he was the guest in the Bill Gates Unconfuse Me podcast, where he talked a lot about the next-generation model. According to Altman, the next-generation model would likely be fully multimodal with support for videos, code, speech, and images. He said they got a great response from images and audio features, and now they will eventually integrate deeper video capabilities.

Altman told Bill Gates that these models will have a steep improvement curve for the next 5 to 10 years. Currently, the AI models are not really reliable. He gave an example to explain it. He said that if we ask GPT-4 most questions 10,000 times, one of those answers will be pretty good. However, GPT-4 often cannot always identify which one is the best. So, that’s the reliability improvement that is important to tackle now.

Besides that, Altman highlighted another notable improvement, i.e., the ability to customize how the AI acts, responds, and solves problems. As for GPT-4, users intend to have different styles of responses along with different assumptions when responding. So, the upcoming models will make that all possible and even support the use of your own data.

In fact, the models are likely to become more capable of knowing about you, your calendar, and your email and also able to connect to outside data. Overall, GPT-5 and the upcoming models are going to improve the shortcomings of the current models and also elevate the capabilities to gradually achieve Artificial General Intelligence (AGI).

What’s the Expected Release Date of GPT-5?

On June 7, 2023, Sam Altman told the Economic Times that they had plenty of work to do prior to GPT-5 and mentioned that they were not close to it. Later, in November 2023, Altman told the Financial Times that GPT-5 is in the early development stage. However, he also mentioned that they need more investment, especially from Microsoft, to build it.

GPT-5 will need more processing power and data than ever before. The training of data alone requires months if not years. On top of that, OpenAI wants to make GPT-5 more reliable and advanced than GPT-4. Therefore, the exact release date of GPT-5 is not confirmed so far.

GPT-5 is currently in the development phase, so it won’t be released anytime soon. Some are anticipating that it will be released in 2024, but nothing is officially confirmed. So, we have to wait for some more time until we get more clarification from OpenAI.

The Pursuit of Artificial General Intelligence (AGI): GPT-4, GPT-5, and the Road Ahead

Most of the world’s largest AI labs, including OpenAI, have Artificial General Intelligence (AGI) as their ultimate goal. It is a form of superintelligence that’s smarter than humanity.

GPT-4 and similar models are impressive AI models that mirror humans in terms of understanding and knowledge. However, companies are aiming to make AI models so capable that they not only surpass humans in terms of understanding but also compete with humans in complex ideas, etc. This is what they are terming as Artificial General Intelligence (AGI), the AI smarter than humanity.

Earlier, it was reported that GPT-5 might be some type of superintelligence, but now there is a buzz that it will be more about improving GPT-4 capabilities. It is expected to have improved reasoning features with some level of understanding beyond simply replicating human intelligence.

Therefore, we can consider GPT-5 a step towards AGI, but there is still a lot of work to be done. In fact, Altman confirmed during the speech at the Y-Combinator W24 that he had told the entrepreneurs and founders in the room to build with the mentality that AGI will be accomplished soon.

Wrapping Up

GPT-4 is already capable of many things we wouldn’t have imagined a few years back. Besides its use in ChatGPT, GPT-4 has become a part of many tools. For instance, UPDF, a top-notch PDF editor tool, is now leveraging GPT-4 to provide an AI chatbot (UPDF AI) to enhance the PDF reading experience.

If GPT-4 is so powerful, the features and capabilities of GPT-5 are just unimaginable right now. According to Sam Altman, GPT-5 will be a big improvement. Until OpenAI provides more technical details about GPT-5, we just have to wait and watch. To wrap up, the upcoming years are definitely going to be full of big advancements in AI, especially towards AGI.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI