GPT-5 is now official as AI enthusiasts have been anticipating since the release of GPT-4. In case you’re a time traveler from the Stone Age, GPT (Generative Pre-trained Transformer) is a large language model (LLM) developed by OpenAI and can perform several human-like tasks. This includes answering questions and summarizing texts by scanning through millions of web articles and books.
You might be wondering when GPT-5 will come out, what cool stuff it can do, and how strong it will be.
Maybe you're wondering if there will be a GPT-4.5 before GPT-5.
And guess what? You're not alone!
Many people are curious about this, including what the big boss of OpenAI, Sam Altman, thinks about it. We're here to give you the scoop on all of these questions and the latest news about GPT-5 that we've been able to discover.
When Will GPT-5 Be Released?
So, when is GPT-5 going to be available? Well, as of now, OpenAI hasn't officially announced the exact date. Back in March, the CEO of Runway, Siqi Chen, tweeted that GPT-5 would finish its training in December.
But in June, Sam Altman clarified that the company is not currently training GPT-5. (We have more on this below.) As for now, it looks like they're still making GPT-4 better and smarter and that's something to look forward to in the next few months.
History and Timeline of GPT 1-4
Before we continue, let’s look back at the brief history and timeline of the GPT series. This will help us understand how OpenAI's language models have improved over time and give us an idea of what GPT-5 might be like.
GPT-1: Introduced in June 2018, this initial iteration of the GPT was trained by BooksCorpus, and contained 117 million parameters. One of its core features is being able to predict the next word in a sentence.
GPT-2: Released in February 2019 but was officially launched in November 2019. This second generation of GPT has 1.5 billion parameters and can generate coherent, multi-paragraph text.
GTP-3: Released in June 2020, this smarter GPT version had a significant leap with a whopping 175 billion parameters. It can now create poems, write computer code, and translate languages. Most importantly, GPT-3 can answer factual questions.
GPT-3.5: Revealed with the launch of ChatGPT, this improved version of GPT 3 has four models, including GPT 3.5 Turbo, which was released on March 1st, 2023. Last August, OpenAI released fine-tuning capabilities for GPT-3.5 Turbo.
ChatGPT: Released on November 30th, 2022, this OpenAI consumer-facing service is specifically designed for chatbot applications. Trained using Reinforcement Learning from Human Feedback (RLHF), this free AI chatbot enjoyed a staggering 1 million users in just 5 days.
GPT-4: Launched on March 14th, 2023, this latest GPT version is exclusively available to ChatGPT Plus users and is believed to have 1.7 trillion parameters! Clearly, GPT-4 is much better than GPT-3.5 at doing complicated things and being reliable.
OpenAI Files for GPT 5 Trademark
Despite having no official release date yet for GPT 5, OpenAI is not taking chances for its exclusive ownership. This is why the AI giant has already applied for the GPT 5 trademark at the United States Patent and Trademark Office.
According to the application, which was filed on July 18, 2023, GPT-5 is a set of downloadable computer programs and downloadable computer software for the following:
- for using language models
- for the artificial production of human speech and text
- for natural language processing, generation, understanding, and analysis
- for machine-learning-based language and speech-processing software
- for the translation of text or speech from one language to another
- for sharing datasets for the purpose of machine learning, predictive analytics, and building language models
- for conversion of audio data files into text
- for voice and speech recognition
- for creating and generating text
- for developing, running, and analyzing algorithms that are able to learn to analyze, classify, and take actions in response to exposure to data
- for developing and implementing artificial neural networks
Despite the move, it appears that GPT 5 will not be released this year as far as AI training is concerned.
While GPT-4 was released merely four months after ChatGPT, OpenAI spent more than two years developing, training, and testing it.
In that case, we may be waiting until 2025 for the release of GPT 5. As for now, we can expect further improvements to GPT-4 (and even a possible GPT -4.5).
Possible Features of GPT-5
Based on the trademark application, we now have some ideas on what GPT 5 can do. But while most of those mentioned are already present in GPT-4, we expect its successor to have a bigger infrastructure. For the time being, here’s what we think are some of the key features that GPT-5 will most likely have:
It’s no secret that GPT-4 experiences “hallucinations,” which is a common complaint among its users. These AI “hallucinations” provide false or misleading information, make errors in reasoning, and cite non-existent sources. To be fair, GPT-4 knows how to apologize for its errors once the user asks for further clarification.
Last May, OpenAI revealed a new strategy they are using to fight hallucinations. The Microsoft-backed AI company calls it Process Supervision, wherein the model is rewarded for each correct step of reasoning rather than the correct outcome (Outcome Supervision). With that, we expect GPT-5 to have a huge reduction in hallucinations which will lead to a higher accuracy rate.
Larger Context Window
The context window refers to the range of words (or tokens for GPT-4) that an AI model can access. The larger the content window a language model has, the more effective it can be in handling complex, lengthier inputs. Consequently, this leads to better performance and increased flexibility in various fields of applications. Needless to say, this breakthrough is a game-changer.
GPT-4 currently has a staggering 32,000 tokens, which equates to around 25,000 words. Given that this is double the 16k tokens for GPT-3.5 Turbo, it’s not a surprise that GPT-4 generates more accurate and contextually relevant responses compared to its predecessor. Therefore, we can expect GPT-5’s context window to be much larger than GPT-4.
Extended Multimodal Capabilities
OpenAI released GPT-4 as a large multimodal AI model, which is a huge leap from GPT-3. Unlike its predecessor which can only interpret text, this fourth-generation language model of GPT can now accept both text and images. It also works with different types and sizes of inputs. Truly, this exceptional feature goes beyond simple conversation with the user and in generating responses.
With the endless possibilities for AI, we can expect extended multimodal capabilities for GPT-5. Apart from text and images, this theorized GPT version may also be capable of working with various media formats including videos. Imagine GPT-5 creating 3D movies based only on an image concept, video samples, or even shared experiences. Better yet, why not include interactive multimedia?
Artificial General Intelligence
Artificial General Intelligence (or AGI) is a hypothetical form of an intelligent agent that allows a language model to achieve unfamiliar tasks. This includes situations that require a higher level of cognitive function.
Simply put, this means AI is able to learn and think like a human, and even better than humans. Interesting but scary, right?
Well, scientists and AI analysts believe that the current language models are now gearing towards AGI, especially GPT-4. In a blog post last February, Altman identified the benefits and risks of AGI to humanity.
Three months later, he testified in a Senate hearing and urged lawmakers to impose tougher regulations on AI systems. So, can we expect GPT-5 to achieve AGI? No, but maybe one step closer. Is that a good thing? Up to you to decide.
ChatGPT is free and uses GPT-3.5, and OpenAI is very unlikely to require users to pay for the service. On the other hand, GPT Plus is a premium service that uses GPT-3.5 Turbo and GPT-4 and costs $20 a month.
For the API, GPT-3.5 Turbo currently costs $0.002 /1K tokens and $0.004/1K tokens for 4K and 16K content, respectively. Meanwhile, GPT-4’s 32K context is priced at $0.12/1K tokens.
Comparatively, the cost of Google’s PaLM 2 is close to that of GPT-4’s API, but the former is only trained on 340 billion parameters. Anthropic AI’s Claude 2 costs $11.2/million tokens or around $0.0465 per thousand words. Note that this AI model has 100,000 context windows (approximately 75,000 words).
So, if OpenAI wants to remain ahead of the race, it should consider making GPT-5 a bit cheaper.
Elon Musk, AI Leaders Call to Pause AI, GPT-5
Through an open letter, billionaire tech mogul Elon Musk, Apple co-founder Steve Wozniak, and other biggest names called on AI giants to pause developments for at least six months. The open letter was published on Future of Life Institute’s website on March 22, 2023, and now has 33,700 signatories. We should recall that Musk co-founded OpenAI in 2015 but stepped down in 2018.
“We call on all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4. This pause should be public and verifiable and include all key actors. If such a pause cannot be enacted quickly, governments should step in and institute a moratorium.” The open letter reads, citing OpenAI’s statement on its vision for AGI.
The ensembled tech leaders noted that “powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable,” and that “this confidence must be well justified and increase with the magnitude of a system's potential effects.” It should be noted, though, that Musk founded an AI startup called xAI, also in March 2023.
OpenAI CEO Sam Altman Talks About GPT-5
The following month, OpenAI CEO Sam Altman was interviewed over Zoom at an event at MIT. You can watch the full video here. But to sum it up, Altman talked briefly about his thoughts on future AI models and categorically denied building GPT-5 at the moment. He was also asked about the open letter, and he responded by defending his company.
“I think we're at the end of the era where it's going to be these, like, giant, giant models, and we'll make them better in other ways. But I would say it, like grows proportionate to the model capacity.” Altman said, suggesting that new AI models may not be about the size but rather on the efficiency.
“There’s parts of the thrust [of the letter] that I really agree with. We spent more than six months after we finished training GPT-4 before we released it. So taking the time to really study the safety model, to get external audits, external red teamers to really try to understand what’s going on and mitigate as much as you can, that’s important,” He said, as his reaction to the open letter.
“I also agree that as capabilities get more and more serious that the safety bar has got to increase. But unfortunately, I think the letter is missing most technical nuance about where we need to pause — an earlier version of the letter claimed we were training GPT-5. We are not and we won’t be for some time,” He added, noting that addressing safety issues is crucial when it comes to AI.
From Altman’s words, we can see an unexpected twist in the race to develop new language models. In a later tweet, Altman reiterated the importance of AGI safety and that frontier models should be regulated. With that, it makes sense to speculate that GPT-5 will not be released sooner than we thought. Nonetheless, we can expect that most of the predicted features will appear in future upgrades.
Will GPT-4.5 Be Released this 2023?
Now that we know GPT-5 is out of the picture (for now), what about GPT-4.5? Well, OpenAI has been silent about it so far, although earlier rumors revealed that GPT-4.5 could be released by September or October 2023.
But since we are now getting closer, it appears that GPT-4.5 is not coming soon. Come to think of it, GPT-3.5 was released around three years after GPT-3.
Nonetheless, note that the situation today and when GPT-3 was released is quite the opposite. Although its release gained more mainstream attention than GPT-2, it was the launch of ChatGPT (and GPT-3.5) that changed everything. And since further fine-tuning and additional features are highly demanded for GPT-4, we will be surprised if GPT-4.5 is coming before the year ends or early 2024.
Is AI Moving Too Fast These Days?
The straight answer is “yes,” and the open letter from tech leaders is solid proof.
And this is not just about GPT, but rather AI-powered chatbots and tools in general. Remember the viral AI image of Donald Trump that was created using Midjourney? The reversed decision of the US Copyright Office involving images also created using Midjourney?
We’ve also seen the negative impact of AI in education once students start to misuse it. How are schools and educators preparing for the challenges this school year to preserve the integrity of education?
Most importantly, the continuous so-called AI anxiety that workers and professionals experience for the fear of being replaced by AI. So yes, I’m happy to know that AI has been “paused” for the past six months.