Introduction of GPT-4o
On May 13, 2024, OpenAI announced a new GPT model, ChatGPT-4o.
GPT-4o has the following improvements compared to previous models:
- Realization of natural conversation
- Faster response time
- Enhanced multilingual support
- Higher reliability
Notably, the improvements in inference speed and quality allow it to perform exceptionally well in real-time dialogue systems, enabling smoother and more natural communication.
Real-Time Dialogue Applications
The following video demonstrates real-time interaction with ChatGPT-4o using a smartphone.
As you can see in the video, ChatGPT-4o can respond at a level close to human conversation.
This allows users to enjoy very natural conversations.
Previously, the ChatGPT series converted voice to text and then inputted it into GPT-4 for dialogue.
This meant that emotional information such as tone was lost in the process.
However, ChatGPT-4o is trained directly from voice, taking into account information such as tone, leading to more natural conversations.
Additionally, it returns responses directly from voice, eliminating the text interpretation step and allowing for faster responses.
Enhanced Multilingual Support
GPT-4o also enhances multilingual support.
Previously, questions asked in English received high-accuracy responses, but questions in Japanese resulted in lower accuracy.
With the enhanced multilingual support, high-accuracy responses can now be obtained even in Japanese.
Summary
This time, we introduced ChatGPT-4o, announced by OpenAI.
The accuracy of the ChatGPT model has remarkably improved, feeling more and more like human conversation.
I believe that how well engineers can utilize generative AI will greatly affect productivity, so I will continue to monitor trends closely.