In June, OpenAI CEO Sam Altman said his company wouldn’t start working on GPT-5 for “some time.”
In mid-July, OpenAI submitted a trademark for GPT-5, hinting that “some time” may be up soon.
OpenAI initiated a trademark application process for “GPT-5” on July 18, 2023. The application was submitted to the United States Patent and Trademark Office (USPTO) by OpenAI OpCo, LLC.
The GPT-5 trademark application, currently under “new application processing,” spans a comprehensive range of categories relating to machine learning (ML) models and AI, including “downloadable software and computer programs engineered for language models, artificial text and speech production, natural language processing, understanding, and analysis.”
The application also covers several other natural language processing (NLP) and text-to-speech sub-disciplines.
You can read the entire list here.
Will GPT-5 encompass artificial general intelligence (AGI)?
OpenAI’s blogs and communications have illustrated a subtle pivot away from just “AI” to “AGI,” – which means “artificial general intelligence.” The company’s “About” page says, “OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.”
Altman and other OpenAI researchers have released blog posts on their plan for AGI and an AGI “charter” published back in 2018.
OpenAI is also committing 20% of its compute resources to AGI R&D and is dedicating resources to AGI alignment, as per a recent blog post.
The “general” in artificial general intelligence denotes AIs that are generally intelligent, e.g., they can perform a wide range of tasks across multiple modalities.
This could, in theory, grant AI models the agency and autonomy they need to meet or exceed humans’ cognitive abilities and skills. However, there is much debate about how, or indeed when, a computer program becomes comparable to humans and other biological organisms that possess a nervous system and are grounded in sensory reality.
Look at it this way – AIs can already perform complex tasks far beyond the human brain’s capabilities, and much faster, too.
On the other hand, a tiny fly can sense the changing air current created by a swatter long before it reaches striking distance and reacts by moving at speeds of hundreds of miles per hour to evade its untimely death. We can’t do that either.
However, of course, there are many things humans can do that other intelligent entities can’t. Humans can compute and act on complex sensory information from all 5 senses in mere milliseconds. Robots couldn’t even catch balls reliably until the mid-2000s.
In the end, systems of intelligence are tough to compare, and intelligent entities of biological and technical classes excel at different things – defining the boundary between AI to AGI will take some serious debate.
Regardless of whether OpenAI’s forecasts for AGI are grounded or realistic, the GPT-5 trademark signifies their commitment to progressing AI technology.
GPT-5 will likely not solely be a large language model (LLM) like ChatGPT but will probably incorporate other tools accessible from the same interface, such as speech and audio synthesis, image generation, and possibly even video generation. The use cases for such a tool are almost unfathomable.
In fact, LLMs in their own right may become a category of models soon to be extinct with the advent of generalist all-in-one models that wrap up language capabilities with several other AI-driven features.
OpenAI might not start working on GPT-5 for “some time,” but given they’re investing heavily in AGI right now, it’s pretty likely that the groundwork is being set, at least.