The company has applied for a new trademark.
OpenAI has applied for new trademark, giving us a tantalizing glimpse of what could be the successor to its blockbuster large language model GPT-4.
The application for the mark “GPT-5,” as spotted by Windows Latest, dates back to July 18, and details a new “downloadable computer software for using language model.”
The wording for the application remains vague, and highlights other features like “the artificial production of human speech and text” and “natural language processing, generation, understanding, and analysis.”
According to the report, that’s in line with the language OpenAI used for its applications for both GPT-3.5 and GPT-4.
Unfortunately, that’s about as much as we can glean from the filing. It doesn’t necessarily indicate when or if OpenAI plans to launch GPT-5, something that’s presumably unlikely to happen any time soon.
New and Shiny
What’s more likely is that OpenAI will focus its efforts on improving its GPT-4 model, instead of pouring all its resources into the next iteration.
The company first revealed GPT-4 back in March and made it generally available via API last month. While it has become tremendously popular, in large part due to OpenAI’s chatbot ChatGPT and integration into Microsoft’s Bing browser, it’s still far from perfect.
The model still tends to hallucinate facts, making it unsuitable for plenty of use cases. It also appears to cost OpenAI tremendous amounts of money to keep running.
While we can only speculate about the company’s next iteration, it’s likely OpenAI will want to address both those pain points. After all, despite paywalling certain GPT-4 features, these models are leaving massive holes in the company’s operating budget.
Whether GPT-5 will be able to plug those holes is anybody’s guess.
More on OpenAI: It’s Impossible for Chatbots to Stop Lying, Experts Say