On August 18, the Washington Post published an article stating,The performance of the much-anticipated new generation GPT-5 large model did not meet the expectations of some people in the technology industry, which caused them to begin to rethink the speed of "superintelligence" and question whether AI can quickly catch up with humans as promoted by some executives.
AI may not catch up with humans so quickly
Before OpenAI made this major upgrade to ChatGPT, technology workers and AI enthusiasts had been looking forward to it for several months.
OpenAI decided to name the new system GPT-5, which has led to comparisons with GPT-4, which will be released in 2023. GPT-4 shocked the technology world and promoted the rapid development of ChatGPT. Currently, the number of weekly users has reached 700 million.
However, after GPT-5 was officially released on August 7 and was reviewed by people in the technology community in the following days, some people in the industry believe that although this upgrade is a welcome improvement, it is not a major leap.
On X, a social platform commonly used in the AI industry, technology commentators posted screenshots showing that GPT-5 performed better than previous AI systems on tasks such as coding or gathering research on complex topics. However, they also shared some evidence thatIt could still make mistakes common to previous chatbots, such as being fooled by puzzles or confidently claiming factually incorrect information as true.
GPT-5 falls short of expectations
The lackluster performance has prompted some tech insiders to recalibrate their expectations for how quickly AI capabilities will improve and question the claims of industry leaders such as OpenAI CEO Sam Altman. These people have said that systems capable of matching human intelligence on most tasks are not far off.
Will AI quickly catch up with humans?
Data shared by OpenAI shows that the upgrade to ChatGPT does make it perform better on many tasks, and it is still expected to transform many workplaces as enterprises use it more widely. But in recent years, AI industry executives have been touting the progress of AI, convincing tech workers, lawmakers and consumers that AI can achieve rapid breakthroughs rather than gradual progress.
Some people who believe that AI is developing extremely fast and may exceed human control within a few years say that GPT-5's performance has allowed them to delay their expected time for this to happen. Others in the tech industry believe thatInstead, this is new evidence that such concerns are unfounded.
OpenAI CEO Altman
David Sacks, President Trump’s head of AI and cryptocurrency business and a well-known technology investor, wrote on his X account that the release of OpenAI is an improvement, butIt also proves that those who predict that head AI systems will quickly surpass other systems and achieve "super intelligence" and surpass humans are wrong.
"Head models perform similarly on performance benchmarks, they still need humans to tell them what to do and verify that they work for them to be truly useful," he wrote. "This is a sign of great engineering and should be celebrated, but this is not what the doomsday rhetoric describes."
expectations too high
Sumuk Shashidhar is an AI researcher and doctoral student at the University of Illinois at Urbana-Champaign who has worked with several AI companies. He said that many people in the industry seem to have too high expectations for OpenAI’s new products.
“For the average user, it feels like there’s not much change.”Shahidhar said. He noted that among those who follow the AI industry closely, there was an expectation that GPT-5 would be like "talking to another very smart person," but that doesn't seem to be the case.
Even some of those who are most concerned that AI advancements may lead to humans losing control of technology believe that GPT-5 has fallen short of expectations.
“People originally thought this was a major technological change,” said Max Winga, a policy analyst at the nonprofit Control AI. The group advocates for stricter rules on AI development to prevent companies from creating super-intelligent technologies that could act autonomously and harm humans.
On the contrary, Wenga believes that the release of GPT-5 "proves that the assumption that there will be some kind of major breakthrough is not tenable." He also noted that he remains concerned about the speed of improvement that AI can achieve, because society is "not ready" for the kind of AI that can learn on its own and become smarter.
Simon Willison, a software engineer and blogger, tested GPT-5 two weeks before its release. He said GPT-5's performance was good, but not surprising.
“It’s not a huge breakthrough over previous versions we’ve used, but it rarely makes mistakes, performs generally competently, and is sometimes impressive at the things I like to do with models,” he wrote in a blog post on the day GPT-5 was released.
Regress?
Some outside the tech community even view GPT-5 as a step back. They complained that the new version was not warm enough compared to previous versions of the conversational experience. The old version was taken offline by OpenAI on the day GPT-5 was released.
OpenAI quickly changed its policy and allowed paying users to continue using the older version of GPT-4o. The company said on Friday it was tweaking GPT-5 to be "warmer and friendlier" based on feedback that users felt it lacked approachability.
Some people believe that GPT-5 has not achieved major improvements compared with previous technologies. OpenAI CEO Altman refuted this statement during a dinner with reporters last Thursday night.
He said that within 48 hours after GPT-5 was released, demand from companies using OpenAI technology in their products doubled, forcing the company to urgently find new computing chips to handle customer inquiries for improved chatbots.
Altman said that to truly understand the latest advances in OpenAI, one needs to work closely with GPT-5 on complex technical issues."This is a model that scientists say has reached the top level of its field in terms of research capabilities."he said.
Altman noted that the technology has limited room for improvement in everyday small talk, so those who primarily use chatbots to discuss their personal lives may not notice GPT-5's enhanced capabilities.
The launch of GPT-5 also shows that OpenAI has begun to pay more attention to financial issues, trying to reduce huge costs while burning more than 30 billion US dollars in venture capital.
This new large model is designed to adjust the use of computing resources based on the complexity of the user's questions, trying to use less computing power for simple questions and more for complex tasks. This could potentially allow OpenAI to increase efficiency by reducing the cost of parts of the conversation.
Altman insisted that the company has not given up on its goal of building more powerful AI and said that it will still bring more breakthroughs in the future, no matter the cost. "We're probably going to spend more money in some areas than any company has ever done," he said.