Getty Images

OpenAI advances LLM with GPT-4o; Google Gemini update looms

The GenAI giant is out with a faster, more interactive version of its GPT-4 LLM for enterprises, and new, higher speed features for the consumer-focused ChatGPT Free.

This story was updated on May 15, 2024.

OpenAI on Monday unveiled GPT-4o, an updated version of its most powerful enterprise-grade large language model, newly engineered to orchestrate quick responses in real time across audio, video and text.

Unveiled in a flashy live demo on YouTube, the newest incarnation of the GPT LLM line conversed naturally and quickly with three OpenAI employees.

It listened to and helped a man solved math equations and visually read his emotions from his facial expression. GPT-4o ("o" for omni) also sang an invented fairy tale in a robot voice and verbally translated a conversation between an Italian and English speaker.

OpenAI also showcased a refreshed user interface for the desktop version of what it called its new flagship model and extended some of GPT-4o's faster, more interactive capabilities to the consumer version of the model, ChatGPT Free.

"GPT-4o provides GPT-4-level intelligence, but it is much faster," said Mira Murati, OpenAI CTO, during the live presentation. "We're really making a huge step forward when it comes to ease of use. And this is incredibly important because we're looking at the future of interaction between ourselves and the machines."

The day after the GPT-4o introduction, OpenAI CEO and co-founder Sam Altman posted on the X platform that chief scientist Ilya Sutskever, another co-founder who has been associated with the “AI safety” faction within OpenAI, is leaving the company.

Altman has been seen as pushing accelerated development of generative AI, with the goal of creating AGI (artificial general intelligence), at the expense of technical design to limit the risks of AI technology. Sutskever’s departure appears to consolidate Altman’s control of OpenAI.

Jan Leike, who ran OpenAI’s safety unit with Sutskever, also resigned. Jakub Pachocki is the new chief scientist.

The GenAI arms race

The release of GPT-4o came on the eve of a similarly splashy product release expected from Google, a competitor to OpenAI and its partner, Microsoft, in the generative AI arena.

The Google I/O developer conference is set to open Tuesday. Observers expect the tech giant and AI pioneer to further ratchet up a high-stakes generative AI arms race that has seen OpenAI, Google, Microsoft, Meta, Amazon and smaller vendors like Anthropic, Cohere and Mistral vie to match and outdo each other on a constant basis over the last two years.

For the first time, you're starting to see OpenAI play catchup to Google.
Chirag DekateAnalyst, Gartner

In this competitive context, OpenAI's achievements with GPT-4o, while technically impressive to a live audience, more realistically simply matched what Google's displayed with its own blockbuster LLM, Gemini, last December, said Gartner analyst Chirag Dekate.

"Any progress in generative AI innovation is always inspiring, because you see researchers on the frontiers of model development and engineering, working really hard to make what seems impossible possible," Dekate said. "But at the same time, frankly speaking, I was underwhelmed."

"For the first time, you're starting to see OpenAI play catchup to Google," he said.

OpenAI versus Google

Both Gemini and GPT-4o are multimodal models, meaning they generate content across text, audio, video and image modalities.

However, Google suffered a serious blow to its public image in February, when Gemini's image generator spat out wildly inaccurate images of people, including Black soldiers in Nazi uniforms.

Google quickly shut off the image generator. Since then, the tech world has waited to see when the vendor would turn it back on and with what safety guardrails to prevent a similar disaster.

As for OpenAI, its Dall-E family of image-generating models, along with ChatGPT and GPT-3.5, were the first LLMs to become a mass phenomenon, in 2021 and 2022.

For some, the performance of OpenAI's generative AI technology on Monday spotlighted the impressive abilities of the fast-evolving technology and its potential to let humans interact with it naturally.

"I think the really impressive part of the OpenAI presentation was them showing that they're doing this live and having the models [hand off] to one another and react pretty close to real time," said William McKeon-White, a Forrester analyst.

McKeon-White noted that OpenAI may have provisioned the demonstration with an unusually large amount of compute power to overcome any delayed responses. Indeed, Murati thanked AI hardware giant Nvidia "for giving us the most advanced studio today."

"But it was still pretty cool to see all of that working in tandem compared to Google's initial announcement -- which was still pretty impressive," McKeon-White said, referring to Google's prerecorded demo. "But this is sort of saying that you can use these yourselves."

However, while OpenAI, a 2015 startup now valued at about $80 billion, and Google boast fairly similar LLM systems, their business approaches are quite different.

Google has added Gemini's generative AI technology into its enterprise cloud-based software, including office productivity and analytics applications and databases. At the same time, it provides its customers with access to many third-party LLMs and foundation models, as well as its own models, in the Model Garden on the Google Vertex AI platform.

Meanwhile, OpenAI, aside from its deal with Microsoft to provide generative AI technology, leases its models to enterprises, other software vendors and individual consumers directly, Dekate noted.

"OpenAI's achievements are impressive in their own right," he said. "But there are limitations to how they can take these to market."

Even so, OpenAI still retains the so-called first-mover advantage, Dekate acknowledged. As the first big player in the mass GenAI market, its steady drumbeat of product releases has set the tone for industry.

OpenAI said it will roll out some GPT-4o capabilities iteratively in the coming weeks, with red team testing starting immediately.

Other GPT-4o text and image capabilities are available in ChatGPT starting now, in the free version and for ChatGPT Plus users, with up to five times higher message limits. The vendor said it will also release in the coming weeks a new version of GPT-4o Voice Mode in in alpha within ChatGPT Plus.

Also, developers can now use GPT-4o in the LLM's API as a text and vision model.

Microsoft Azure OpenAI Service customers can test GPT-4o's capabilities using a preview playground in Azure OpenAI Studio starting today. 

Shaun Sutner is senior news director for TechTarget Editorial's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. He is a veteran journalist with more than 30 years of news experience.

Dig Deeper on AI technologies

Business Analytics
CIO
Data Management
ERP
Close