SAN FRANCISCO, August 7
OpenAI unveiled its GPT-5 artificial intelligence model on Thursday, marking a much-anticipated advancement in technology that aims to reshape global business and culture.
The GPT model, which powers the popular ChatGPT chatbot, is now available to all 700 million ChatGPT users, according to OpenAI.
A significant concern is whether companies that launched the generative AI trend can sustain meaningful technological advancements that attract corporate users, thereby justifying the hefty sums spent on these developments.
This release comes at a pivotal moment for the AI sector. Major players like OpenAI, Alphabet, Meta, Amazon, and Microsoft are ramping up investments in AI data centers, anticipating a total spend of nearly $400 billion this fiscal year.
OpenAI is having early discussions about enabling employees to cash out based on a valuation of $500 billion, a notable increase from the current $300 billion valuation. Leading AI researchers are reportedly commanding signature bonuses of around $100 million.
“Business spending on AI has been pretty weak so far, but consumer spending on AI has been quite strong. Personally, I really enjoy chatting with ChatGPT,” remarked economics writer Noah Smith. “But, consumer spending alone isn’t enough to justify all the investments in AI data centers.”
OpenAI emphasizes the capabilities of GPT-5, stating that it excels in software development, writing, health-related inquiries, and finance.
“This is the first instance where I feel one of our main models can genuinely engage with experts, even PhD-level specialists,” Sam Altman mentioned during a press conference.
He added, “One of the coolest applications is generating quality software instantly.”
A key measure of success will be how the progression from GPT-4 to GPT-5 compares to earlier advancements. Early reviewers have indicated that while GPT-5 impresses in coding and solving scientific and mathematical issues, the change from GPT-4 isn’t as significant as seen previously.
Reflecting on Progress
Almost three years ago, ChatGPT introduced engaging interactions with human-like writing, quickly emerging as one of the fastest-growing applications.
In March 2023, OpenAI launched GPT-4, a substantial leap in intelligence compared to its predecessor, GPT-3.5, which had scored in the bottom 10% on a bar exam. In contrast, GPT-4 achieved a spot in the top 10% for the simulated exam.
The leap to GPT-4 was attributed to enhanced computing power and data, and similarly, there’s hope that increasing scale will lead to further improvements in AI models.
However, OpenAI faces challenges, particularly concerning data limitations. Ilya Sutskever, the former chief scientist at OpenAI, noted that while processing power is on the rise, the quantity of data has remained stagnant.
He highlighted that these large language models rely on extensive datasets derived from the internet, with limited alternatives for sourcing human-generated text data.
Alongside data scarcity, the complexity of the system could lead to hardware failures during extensive model training runs, leaving researchers uncertain about the model’s performance until the end.
OpenAI is exploring a novel approach known as “test time calculations,” which allows AI models to take more time considering each question, enhancing their problem-solving capabilities in areas like mathematics and complex reasoning.
In this context, GPT-5 operates as a type of router. When faced with challenging inquiries, it employs test time computing to generate responses.
This marks the first instance where the public can access OpenAI’s test time calculation technology, which Altman stresses is crucial to the company’s mission of creating AI that benefits humanity as a whole.
Altman believes that investments in AI still fall short of what’s necessary for global infrastructure development to make AI accessible in various markets.
