GPT-5 Will Have Ph.D.-Level Intelligence, OpenAI’s CTO Says

In a recent interview hosted at Dartmouth Engineering with Jeff Blackburn, former Vice President of Global Media and Entertainment at Amazon, OpenAI’s CTO Mira Murati revealed the GPT-5 model, which was expected to be released this year, will be launched in late 2025 or early 2026.

Key Points:

Murati described the upcoming GPT-5 model as having a PhD level of intelligence. “If you look at the trajectory of improvements of AI systems like GPT-3 were, maybe, let’s say toddler level intelligence, systems like GPT-4 are more like smart high schooler intelligence. And then in the next couple of years, we’re looking at PhD level intelligence,” she explained.

GPT-5 having PhD level intelligence seems to be an extremely powerful AI tool that’s gonna impact every field you can imagine. It could help pass university exams, create film and TV scripts, give expert advice in finance, healthcare, and law, design and deliver customized learning experiences for an individual, produce high-quality technical documents and research papers, and design sophisticated software architectures.

“With GPT-5, the research part of the work can be done much faster … it’s going to be a collaborative tool, especially in the creative spaces….it’ll have a new layer of productivity,” Mira Murati said. GPT-5 will be a combination of AI tools like analysis and research tools, and it will work seamlessly deploying different advanced tools right at the moment when they’re required to be used. 

Will GPT-5 Take Away Human Jobs?

Since GPT-5 is an extremely intelligent AI generative model, it’s a little bit scary too in terms of its risk of taking over many human jobs. Mira Murati herself anticipated “a lot of jobs will change, some jobs will be lost, some jobs will be gained.”

This upcoming GPT can be expected to take over the strictly repetitive nature of jobs where you don’t have any chances of further improvement. These repetitive tasks include testing code, data entry, making repetitive sales or promotional calls, providing basic customer service, managing financial transactions, inspecting products for defects in a standardized manner, arranging and managing appointments and schedules, digitizing physical documents and categorizing them, and other cognitive labor/works.

People have been discussing the job displacement, since the launch of GPT-3, as an outcome of advancement in artificial intelligence. It’s a reality, indeed. AI is really taking away some kinds of jobs, especially repetitive ones as Murati said. But it’s also creating new kinds of jobs especially where AI is involved. So, I believe it’s better to learn the use of AI in your own field to get better at what you do and thus to avoid losing your job.

What Will Be The Good Impact of GPT-5 On Different Industries?

GPT-5, with PhD-level intelligence, can significantly impact various industries in several positive ways. In education, it’ll have a huge impact. It will provide a highly personalized tutoring and learning experience for an individual. It would make quality education accessible for everyone. Students residing in underdeveloped areas could get advanced knowledge of things.

Similarly, in healthcare, it’ll enhance diagnostics and provide personalized treatment plans by analyzing the medical data of an individual. In finance, it will improve financial forecasting, risk assessment, and fraud detection through advanced data analysis. Additionally, it’ll offer personalized financial advice.

The list of fields where it can impact positively is endless. Because of its PhD level of intelligence, and advanced analysis and research capabilities, it will help in legal research, enhancing customer support through intelligent chatbots capable of handling complex queries, providing deeper insights into consumer behavior, optimizing ad targeting, and generating high-quality content.

Are There Any Safety Risks Attached To The Use Or Misuse of GPT-5?

On Jeff Blackburn’s question about safety concerns regarding the smart AI agents, like agents based on GPT-5, connecting to the internet and performing different tasks on their own will pose a threat to humans or not, Murati replied with an analogy: “It’s like training a smart dog versus a dumber dog.” She suggested that the more intelligent a GPT model is, the easier it is to train that model to avoid doing certain tasks that could raise safety concerns against humans.

She also revealed that a certain feature of any AI model by OpenAI can be downsized by the company because of its misuse. As in the recent controversy around the voice of Sky used as a voice assistance feature of the GPT-4o model, they took this voice down as some people found it similar to the voice of Scarlett Johansson. However, Mira declared that it was not her voice, it just had some similarities. 

“OpenAI doesn’t allow people to make their own voices with this technology,” She said. So, the company made the feature of voice assistance less capable in order to avoid its misuse.

I think she clarified enough that the misuse of AI can’t be avoided 100% but it’s easy to train a smarter AI model to reject doing certain tasks. And that intelligence and safety go hand in hand. If an AI tool is smarter, it understands the guardrails better.

Will GPT-5 Be Regulated By Government Before Its Release?

Regulation balances the benefits of AI with the need to protect individuals and society from potential risks. AI models need government regulation to ensure safety, security, and ethical standards, preventing harmful or biased applications. It establishes accountability, safeguards privacy, and prevents misuse, such as in deep fakes and autonomous weapons. 

Additionally, regulation ensures market stability by fostering competition and innovation while building public trust through transparent oversight.

In the interview, Blackburn asked Murati: “You probably don’t want Washington D.C. regulating your release of GPT-5.” She answered that although more and more people having knowledge of AI are joining the government, they’re not enough yet to enable the government to regulate it.

She revealed that OpenAI has already regulated its AI models and that regulations “ actually informed White House commitments or what the UN Commission is doing with the principles for AI deployments.”

She clarified that in other fields, regulations are made after a work is done or a product is released whereas in the field of AI, regulations are made before an AI system is launched, based on “forecasting and science of capability prediction”. So, governments need to have such people who can do so.