The GenAI giant presents a faster and more interactive version of GPT-4. This newly released LLM is aimed at businesses, but its new features will also be accessible in ChatGPT, including the free version.
OpenAI has just unveiled GPT-4o, an updated version of GPT-4, its most powerful large language model (LLM) targeting businesses. GPT-4o has been redesigned to offer faster (real-time) responses whether using audio, video, or text.
Unveiled during a live demonstration on YouTube, the LLM conversed naturally with three OpenAI employees. It listened to and helped a man solve mathematical equations and read his emotions from his facial expressions. GPT-4o ("o" for "omni") also sang a fairy tale it invented, with a synthesized voice. Then it orally translated a conversation between an Italian and an Englishman.
OpenAI also unveiled a refreshed user interface for the desktop version, which is now its new flagship model. Certain text and image features of GPT-4o are now available in ChatGPT, on the free version and for ChatGPT Plus users, with message limits up to five times higher for the latter.
The publisher indicated that it would also release a new version of the GPT-4o voice mode in alpha for ChatGPT Plus in the coming weeks, and that Read Teaming would begin immediately.
Developers can now use GPT-4o in the LLM API as a text and vision model. And Microsoft Azure OpenAI Services customers can test GPT-4o's capabilities in a preview playground in Azure OpenAI Studio starting today.
"GPT-4o offers intelligence on par with GPT-4, but it is much faster," summarizes Mira Murati, CTO of OpenAI. "We're taking a big step forward in terms of ergonomics. And this is very important, because we are working on how, in the future, we will interact with machines."
A Highly Competitive Sector
The release of GPT-4o took place the day before another highly anticipated event: the Google I/O developer conference.
During this event, observers expect Google to announce new features for its Gemini offering.
The GenAI sector is already very competitive with OpenAI and Google, but also with Microsoft, Meta, Amazon, and smaller players like Anthropic, Cohere, or the European companies Mistral, Aleph Alpha, and LightOn.
Google I/O is expected to further intensify this fierce competition. In this context of rivalry between OpenAI and Google, OpenAI's new GPT-4o LLM, while technically impressive, would only match what Google showed with Gemini in December, believes Chirag Dekate, an analyst at Gartner.
"Any advance in generative AI innovation is always inspiring. We see researchers at the forefront of model development and engineering working very hard to make what seemed impossible possible," he explains. "But at the same time, frankly, I was disappointed."
"For the first time, it's OpenAI that's starting to catch up with Google," he adds.
Forrester has a more nuanced opinion. "I think the most impressive part of OpenAI's presentation is that they showed models communicating with each other and reacting almost in real-time. And they showed it live," boasts William McKeon-White, an analyst at Forrester.
OpenAI certainly mobilized enormous amounts of computing power for this demo. "But it was still pretty cool [...] compared to Google's demo," continues the Forrester expert, referring to Google's pre-recorded demo.
OpenAI vs Google
Gemini and GPT-4o are both multimodal models, meaning they generate content from text, sound, video, and images.
In February, Google suffered a serious setback to its image and a false start. Its image generator had indeed released images of completely incoherent people - like black Nazi soldiers. Google quickly shut down its tool.
On OpenAI's side, Dall-E, ChatGPT, and GPT-3.5 were the first LLMs to become a mass phenomenon in 2021 and 2022. The advantage seemed to be on the side of Microsoft's protégé. But technology is evolving fast. Very fast. And while OpenAI is today valued at around $80 billion, Google can boast of having fairly similar LLMs.
Their approaches remain very different, however.
Google has integrated its GenAI into its cloud solutions: office productivity suite, analytics, and databases. At the same time, it allows its customers to access its models and many third-party LLMs and foundation models in the "Model Garden" of its Vertex AI platform.
For its part, in addition to its agreement with Microsoft to offer its generative AI technologies on Azure, OpenAI markets its models directly to businesses, other publishers, and individuals, compares Chirag Dekate.
"The results achieved by OpenAI are impressive in themselves," he concedes. "But there are limitations to how they bring them to market." The Gartner expert nevertheless notes that OpenAI retains the "first mover" advantage, even if Google is catching up.
Post a Comment