ChatGPT has become a sensation in the tech world by showcasing artificial intelligence (AI) with conversational abilities that surpass anything we have ever seen before. The viral chatbot interface relies on GPT-3, which is believed to be one of the largest and most complex language models ever developed. It has been trained on a staggering 175 billion data points, known as “parameters”.
OpenAI, the AI research organization behind GPT-3, is reportedly working on a successor called GPT-4. According to rumors, GPT-4 is expected to be even more powerful and capable than GPT-3, with some sources suggesting a parameter count of around 100 trillion. However, the exact size of GPT-4 remains unconfirmed, and Sam Altman, CEO of OpenAI, has disputed some of these rumors in colorful language.
There is speculation that GPT-4 may already be in use, but nothing has been confirmed. Some believe that the new ChatGPT feature in Microsoft’s Bing search engine could be powered by GPT-4, despite no official announcement. This theory is plausible due to Microsoft’s recent $10 billion investment in OpenAI, making them the largest single shareholder.
What sets GPT-4 apart from its predecessors? Can it bring us closer to the vision shared by Google CEO Sundar Pichai, of AI having an impact on society that is “more profound than fire or electricity”? Here’s what we know about GPT-4 at this point in time:
Release of GPT-4 is expected in 2023
Currently, there is no official confirmation about the release of GPT-4. However, rumors have been circulating in the tech industry, including reports from outlets such as the New York Times. Some speculate that GPT-4 may already be in use in Bing’s chat functionality, which is powered by ChatGPT.
Although access to ChatGPT-powered Bing is currently limited to those on a waiting list, Microsoft plans to open it to millions of users by the end of February. If these rumors are untrue and Bing is running on GPT-3 or an updated version, GPT-3.5, we may have to wait longer for GPT-4. Like GPT-3, GPT-4 may be released first to select partners, paying customers, and academic institutions before becoming widely available to the public.
GPT-4 may not have access to significantly more training data than GPT-3
It is uncertain at this time, but it is believed that GPT-4 will be much larger than GPT-3, potentially by up to 100 times. Some sources claim that it could have around 17 trillion parameters, although OpenAI’s CEO, Sam Altman, has called the notion that it will be trained on 100 trillion parameters “complete bullshit.” Nonetheless, Altman has also suggested that the focus may not be solely on increasing the model’s size, but rather on improving its ability to use existing data efficiently. This could be a wise move, given that a competitor language model, Megatron 3, is trained on more data than GPT-3 but doesn’t outperform it. Thus, it’s not always the case that bigger is better in the world of AI. Improving efficiency would not only reduce the running cost of GPT-4 but also of its spin-off, ChatGPT. This will be a crucial factor if GPT-4 is to be as widely used as the most popular search engines, as some experts predict.
GPT-4 is expected to excel in generating computer code
ChatGPT, and its underlying technology GPT-3, has a remarkable ability to generate both human and computer languages. This means that it can create computer code in various programming languages, such as Python, C++, and Javascript, which are widely used in software development, data analytics, and web development.
Recently, OpenAI has been on the lookout for programmers and developers with expertise in describing their code using natural language. This development has led many to predict that future products, such as GPT-4, will push the boundaries even further in generating computer code. This advancement could result in more powerful tools like Microsoft’s Github Copilot, which currently uses a fine-tuned version of GPT-3 to enhance its ability to translate natural language into code.
GPT-4 is not expected to include graphics among its capabilities
Some experts had been speculating that the next advancement in generative AI would involve a fusion of GPT-3’s text generation with the image creation capabilities of OpenAI’s Dall-E 2. This idea is thrilling because it opens up the possibility of generating charts, graphics, and other visualizations which GPT-3 is currently incapable of. However, Altman, the CEO of OpenAI, denied this claim and stated that GPT-4 would remain solely a text-based model.
GPT-4 is expected to disappoint some people
GPT-3 has generated a lot of excitement, and this raises the question of whether future iterations, such as GPT-4, will be as groundbreaking. It’s understandable to wonder whether we’ll still be amazed by a computer writing slightly better poetry after already being amazed by one writing poetry in the first place. Even Sam Altman, a co-founder of OpenAI, has expressed this sentiment, saying in a January interview, “The GPT-4 rumor mill is a ridiculous thing. I don’t know where it all comes from … people are begging to be disappointed, and they will be.”
