Francis Lelong CEO of Alegria.group gives his analysis of the OpenAI announcements of May 2024

I'm delighted to announce a major breakthrough in the field of Artificial Intelligence, marking significant progress towards perfect interaction with technology! Yes, I'm talking about OpenAI's new multimodal model, GPT-4o, a major step towards more natural and fluid man-machine interaction than ever before.

GPT-4o truly embodies the convergence of technologies to deliver an unprecedented user experience. Imagine being able to communicate with a computer system using any medium: text, audio, image or video, and getting responses in the same format, whether text, audio or image. It's now possible, thanks to OpenAI and their incredible innovation, GPT-4o.

What makes GPT-4o truly exceptional is its ability to understand and respond to audio inputs in record time, as fast as 232 milliseconds, with an average of 320 milliseconds. This level of responsiveness is comparable to the human response time in a conversation, and we're obviously thinking here of all the possibilities this opens up in the call center field, for example.

But that's not all. GPT-4o also offers outstanding performance in natural language comprehension and vision. It matches the GPT-4 Turbo's performance on English text and code, while offering a significant improvement on text in other languages, including of course French. Having done a few tests since yesterday, it's really very fast! What's more, it's 50% cheaper in terms of API usage, which is no mean feat when you consider the resource issues involved in using AI on a large scale.

To achieve these impressive results, OpenAI has fundamentally rethought the way AI systems process data. With GPT-4o, they have trained a single model across all modalities, from text to vision to audio. This means that all inputs and outputs are processed by the same neural network, eliminating information loss and enabling richer, more contextual interactions than ever before.

At this stage, data security is at least as important as performance, so OpenAi didn't miss the opportunity to improve its tool in this area either. This is why GPT-4o incorporates end-to-end security mechanisms, from filtering training data to refining model behavior after training. OpenAI has also implemented new security systems to frame audio output, guaranteeing a safe and reliable user experience. Preventing deep fakes will undoubtedly be a crucial issue in the coming months.

All in all, it's a fine evolution of GPT that OpenAI is offering us here, to further boost competition with Google, which was also holding its big conference this week, and the other models. Interaction will be even smoother, and we'll no doubt be one step closer to the AI that dialogues with Joaquin Phoenix in the famous "HER". What a pleasure it is to discover with you every week this AI revolution that is being built before our very eyes!

Francis Lelong CEO, Alegria.group