GPT-5, developed by OpenAI, will be released soon. Customers and developers at OpenAI are telling us what to expect.
The AI arms race is still going strong, with OpenAI vying to develop the largest, baddest model versus Anthropic, Meta, and a revitalized Google. After OpenAI released GPT-4, its rivals had to run to catch up, with some of them coming quite close.
GPT-5 is a new variable in this battle that has the potential to completely destroy the opposition or prove to be completely pointless.
Working diligently on its most recent model, OpenAI hopes it will herald the kind of revolutionary paradigm shift that caught the public’s attention when ChatGPT was released back in 2022.
In a recent research note, Enders Analysis analysts Hamish Low and others said, “OpenAI will be transformative for its own prospects, but also the economy more broadly, if it can deliver technology that matches its ambitious vision for what AI can be.” “Falling short could be fatal.”
For OpenAI, which is up against an increasing number of well-funded competitors, the stakes are high. The analysts went on to say that the business needed to maintain its leadership in AI in order to convince its major tech funders of its viability.
In what ways will GPT-5 be superior?
GPT-5 will undoubtedly have greater power than GPT-4. Will it, however, be sufficient to distinguish OpenAI—which looks to be losing billions of dollars—from the growingly remarkable selection of other AI models available?
Not all of the prospective buyers of GPT-5 are convinced.
CEO and cofounder of Casetext, an AI-powered legal assistant that Thomson Reuters recently bought, said, “I don’t know if it’s going to feel as big.”
Heller is already using GPT-4 and other commercial models to assist attorneys with a plethora of activities that would typically be performed by $500-an-hour first-year associates, such as reviewing contracts and poring over information.
Heller compared the transition from GPT-3 to GPT-4 to going from “grade school to undergrad,” a tremendous, game-changing transformation. He anticipates GPT-5 to be more like transitioning from undergrad to a PhD program—better, but not necessarily game-changing.
Multiple-phase tasks
Heller’s main ambition for GPT-5 is that it will be able to “take more agentic actions,” that is, finish tasks involving several difficult phases without getting off track. This could entail reading a legal brief, researching the appropriate statute, cross-referencing the case law, comparing it to the evidence, and then developing a question for a deposition.
“Right now, I’d say the models aren’t quite clever enough,” Heller stated. “You see sometimes it kind of gets stuck or just veers off in the wrong direction.”
To someone like Heller, GPT-5 could replace not just a first-year legal associate but also a third-year associate or perhaps a partner if it can be relied upon to function independently and make sound decisions about how to handle multi-phase jobs.
“Right now, I’d say the models aren’t quite clever enough,” Heller stated. “You see sometimes it kind of gets stuck or just veers off in the wrong direction.”
To someone like Heller, GPT-5 could replace not just a first-year legal associate but also a third-year associate or perhaps a partner if it can be relied upon to function independently and make sound decisions about how to handle multi-phase jobs.
An expanded context window
According to Heller, he did anticipate that the new model will have a much wider context window. This would enable it to compare contracts or other legal documents that may run into the hundreds of pages at a time and handle larger blocks of text at once.
The possibility that GPT-5 will be able to operate with text, video, and audio interchangeably excites him as well.
“Just yesterday, I was speaking with a judge who asked, ‘When will we be able to review the CCTV or police bodycam footage from the crime scene and use it as evidence?'” Thus, consumers are undoubtedly requesting it,” Heller stated.
Will the cost of GPT-5 be justified?
The technology of GPT-5 is expected to be superior, as most people agree, but the less glamorous but equally significant concern is whether all these new features will be worth the extra money.
“Unless I am extremely confident that the economics will make sense, I don’t want to make that investment,” stated Hooman Radfar, CEO of Collective, an AI-powered platform for independent contractors. AI is used by Collective for tasks like tax implications analysis and business expense classification.
He claimed that he continually compared his in-house systems against commercially accessible AI products to determine when to train models internally and when to purchase pre-made ones. He claimed that Collective’s own models beat GPT-4 by up to 40% on a number of tasks.
However, Radfar is looking forward to GPT-5 because he believes it will have better reasoning skills and be able to provide his customers with answers to their difficult queries as well as an explanation of how it arrived at those answers—a crucial distinction.
