On August 7, OpenAI unveiled its latest artificial intelligence model, GPT-5, generating excitement about the new capabilities that could reshape business and cultural landscapes around the globe.
The GPT series serves as the foundational technology behind the widely used ChatGPT chatbot, and OpenAI announced that GPT-5 will be accessible to all 700 million users of ChatGPT.
Discover more about OpenAI’s updated shopping functionalities incorporated into ChatGPT
The crucial inquiry now is whether OpenAI, the pioneer of the generative AI movement, can sustain the momentum necessary for substantial technological progress that would attract enterprise clients and justify the large investments it is making to advance this technology.
This launch arrives at a pivotal moment for the AI sector, as major players like Alphabet, Meta, Amazon, and Microsoft—who supports OpenAI—are significantly ramping up their spending to establish AI data centers, thereby bolstering investor optimism about future returns. Collectively, these four giants plan to invest almost $400 billion in the current fiscal year.
Currently, OpenAI is reportedly in preliminary talks to enable its employees to cash out at a valuation of $500 billion, up from its present valuation of $300 billion. Top-tier AI researchers are now seeing signing bonuses reaching $100 million.
According to economics commentator Noah Smith, “Business investment in AI has generally been somewhat lacking, while consumer expenditure has been relatively strong due to the popularity of engaging with ChatGPT. However, the consumer spending on AI likely won’t be enough to account for all the investments directed towards AI data centers.”
OpenAI is focusing on the enterprise capabilities of GPT-5, highlighting its strengths in software development, writing assistance, medical inquiries, and financial tasks.
OpenAI CEO Sam Altman remarked during a press conference, “With GPT-5, we finally have a model that resembles a legitimate expert—similar to a PhD—which you can consult on various topics.”
He highlighted, “A particularly exciting feature is its capacity to generate quality software instantly. The concept of on-demand software generation will become a hallmark of the GPT-5 era.”
During demonstrations, OpenAI showcased how GPT-5 could produce fully functional pieces of software based solely on textual prompts, a practice referred to as “vibe coding.”
A critical aspect of assessing success will be whether the advancements from GPT-4 to GPT-5 are comparable to OpenAI’s previous leaps. Two early testers informed Reuters that while they were impressed by GPT-5’s coding capabilities and its proficiency in tackling scientific and mathematical questions, they felt the progression from GPT-4 to GPT-5 wasn’t as substantial as earlier advancements.
Even if the enhancements are significant, GPT-5 isn’t sophisticated enough to entirely replace human roles. Altman stated that GPT-5 still lacks autonomous learning abilities, which is crucial for AI to replicate human skills.
On his well-known AI podcast, Dwarkesh Patel compared the current state of AI to teaching a child to play saxophone solely based on notes from a previous learner.
“If a student attempts it once, a mistake prompts them being sent away to receive detailed feedback on what went wrong. The next student then tries to follow those notes, but this method is flawed in ensuring proper learning,” he commented.
Further Reflections
Nearly three years back, ChatGPT introduced generative AI to a global audience, captivating users with its ability to produce human-like text and poetry, rapidly becoming one of the most swiftly adopted applications in history.
In March 2023, OpenAI released GPT-4, an advanced language model that achieved significant improvement in cognitive performance compared to its predecessor. While GPT-3.5 scored in the lower 10 percent on a bar exam, GPT-4 excelled, placing in the top 10 percent.
The progress of GPT-4 was attributed to enhanced computational resources and data, and the company aimed to achieve consistent improvements in AI models by applying similar scaling strategies.
However, OpenAI faced challenges with scaling. One major issue was the limitation of available data, as noted by former chief scientist Ilya Sutskever, who pointed out that even though processing capabilities were increasing, the data volume was not keeping pace.
Sutskever explained that large language models rely on vast datasets sourced from the internet, and AI research facilities face constraints in acquiring extensive collections of human-created text.
Alongside data limitations, the “training runs” for expansive models are prone to hardware-related failures, given the complexity of the system, with performance insights often only available after lengthy training periods.
Meanwhile, OpenAI found a new avenue for enhancing AI capability, termed “test-time compute.” This approach allows the AI model to dedicate additional compute resources to “ponder” over each inquiry, enabling it to tackle challenging problems requiring advanced reasoning and decision-making skills.
With GPT-5, if a user presents a particularly difficult question, it will utilize test-time compute to generate a response.
This marks the first occasion when the general public will have access to OpenAI’s test-time compute feature, which Altman considers vital to the company’s ambition of developing AI technologies that are beneficial for all of humanity.
Altman also emphasizes the need for further investment: “We must create significantly more infrastructure on a global scale to ensure that AI is readily accessible in various markets.”