OpenAI is reportedly gearing up to release a more powerful version of ChatGPT in the coming months.
The new AI model, known as GPT-5, is scheduled for release this summer, according to two familiar sources who spoke to Business Insider. Ahead of the launch, some companies reportedly tried out a demo of the tool, allowing them to test its improved capabilities.
The technology is part of OpenAI's futuristic quest for artificial general intelligence (AGI), or systems that are smarter than humans.
This lofty, sci-fi premise predicts an AI that can think for itself and therefore create more AI models of its kind without the need for human supervision. Depending on who you ask, such a breakthrough could destroy the world or boost it.
As the race to build the best AI heats up, here's everything you need to know about GPT-5.
What is GPT-5?
GPT-5 is the successor to GPT-4, OpenAI's fourth-generation chatbot that requires you to pay a monthly fee to use.
GPT stands for generative pre-trained transformer, an AI engine built and refined by OpenAI to power the various versions of ChatGPT. Just like the processor in your computer, each new edition of the chatbot runs on a brand new GPT with more features.
In the case of GPT-4, the AI chatbot can provide human-like responses and even recognize and generate images and speech. The successor, GPT-5, will reportedly offer better personalization, fewer errors, and handle more types of content, eventually including video.
Others, such as Google and Meta, have released their own GPTs with their own names, all of which are collectively known as major language models.
These AI systems are based on the human brain and have the ability to generate text as part of a conversation.
Most interesting part of @sama talk: GPT5 and GPT6 are "in the bag", but that's probably NOT AGI (e.g. something that can solve quantum gravity) without some breakthroughs in reasoning. Strongly agree.
- Omar Shams (@smahsramo) September 24, 2023
Is GPT-5 trained?
The latest report claims that OpenAI has started training GPT-5 as it prepares to release the AI model mid-year. Once the training is complete, the system will go through multiple stages of security testing, according to Business Insider.
The story continues
As part of this process, the bot will be "red teamed," a technique in which internal and external testers put the bot to the test and provide feedback on its strengths and weaknesses.
The report follows speculation that GPT-5's learning process has recently begun, based on a recent tweet from an OpenAI official.
There's no adrenaline rush like starting a massive GPU workout
- Jason Wei (@_jasonwei) January 25, 2024
In January, one of the tech company's leading researchers hinted that OpenAI was training a much larger GPU than usual. The revelation followed a separate tweet from OpenAI's co-founder and president, detailing how the company had expanded its computing resources.
A GPU, short for graphics processing unit, is like a calculator that helps an AI model work out the connections between different types of data, such as associating an image with its textual description.
Building at OpenAI is an exercise in maximizing every available computing resource, scientifically predicting and understanding the resulting systems, searching for new or old ideas that are ready to work now, and scaling beyond precedents.
- Greg Brockman (@gdb) January 25, 2024
For his part, Mr. Altman confirmed that his company worked on GPT-5 at least twice last fall.
The first of these was during a talk at the alumni reunion of his former venture capital firm Y Combinator last September, according to two people who attended the event. Mr. Altman said GPT-5 and its successor, GPT-6, were "in the bag" and superior to their predecessors.
In November, he made its existence public, telling the Financial Times that OpenAI was working on GPT-5, although he did not reveal the release date.
More recently, a report claimed that OpenAI's boss had hatched a bold plan to purchase the massive quantities of GPUs needed to train larger AI models.
To overcome supply shortages that are hampering technological innovation, Altman wants to raise as much as $7 trillion to accelerate chip production with the help of a global network of investors, governments and energy suppliers, according to the Wall Street Journal.
What can GPT-5 do?
In February, OpenAI's chief spoke about GPT-5 at the World Governments Summit in Dubai. He said the next version of ChatGPT will be smarter than its predecessors.
"This is more important than it sounds, because what makes these models so magical is that they are generic," Altman explains. "So if it's a little smarter, it's a little better at everything."
Referring to brain power, Mr Altman told the FT that GPT-5 would need more data to train on. The plan, he said, was to use publicly available datasets from the Internet, along with large-scale organizations' own datasets. The last of these involves long-form writing or conversations in any format.
Speaking on Bill Gates' Unconfuse Me podcast in mid-January, Mr Altman said: "Multimodality will certainly be important. That means speech in, speech out. Pictures. Finally video. It's clear that people really want that. We will be able to take that much further, but perhaps the most important areas of progress will be in reasoning."
He continued: "Right now, GPT-4 can only reason in very limited ways. Also reliability. If you ask most questions to GPT-4 10,000 times, one of those 10,000 is probably pretty good, but it doesn't always know which one. and you want to get the best response of 10,000 every time, so increased reliability is going to be important."
GPT-5 vs GPT-4
How can it beat GPT-4? Above all, it will have to surpass GPT-4 Turbo, the next-generation model that OpenAI released to paying subscribers in November.
The company's most advanced AI chatbot has knowledge of world events through April 2023, compared to 2021 for GPT-4; it can even analyze longer prompts of up to 128,000 tokens, or about the length of a 300-page book; it is better at following instructions; and it can automatically switch between tools, including the Dall-E 3 image generator and the Bing search engine, based on user requests.
Both OpenAI and various researchers have also tested the chatbot on real-life exams. GPT-4 proved to have a decent chance of passing the difficult Chartered Financial Analyst (CFA) exam. It scored in the 90th percentile on the bar exam, passed the SAT reading and writing section, and was in the 99th to 100th percentile on the 2020 US Biology Olympiad semifinal exam.
When will GPT-5 be available?
At the time of writing, OpenAI has not yet announced a launch date for GPT-5.
It's also unclear whether the company was affected by the turmoil at OpenAI late last year. On November 17, Mr. Altman was ousted by the company's board of directors. After five days of tumult symptomatic of the dueling positions on the future of AI, Mr. Altman was back at the helm, along with a new board.
Strangely, some ChatGPT users previously claimed that the bot told them it was running on a new AI model called GPT-4.5 Turbo, but that turned out to be a bug.
If OpenAI's GPT release timeline tells us anything, it's that the gap between updates is getting shorter and shorter. GPT-1 arrived in June 2018, followed by GPT-2 in February 2019, then GPT-3 in June 2020, and the current free version of ChatGPT (GPT 3.5) in December 2022, while GPT-4 arrived just three months later. March 2023. Updates have also been received more frequently in recent months, including a 'turbo' version of the bot.
So could June be the magical month again? Or will OpenAI wait until the Developer Day conference, likely in November, to announce GPT-5? We'll have to wait and see.
