Hands-on with the Faster, Smarter ChatGPT-4o AI

OpenAI is upgrading its signature AI system with a new model offering faster responses, more natural language abilities and better integration of text, images, and data. The new model is called ChatGPT-4o, with the "o" standing for "omni." 

In my initial hands-on time with GPT-4o, I found its biggest immediate benefits are much faster response times and the ability to hold a more natural voice conversation. That means you can speak to it directly, have it answer in real time (much like Alexa or Siri might), and even better, you can interrupt the answer and redirect the conversation, all via voice. Voice chat was available previously in ChatGPT, but it felt like much more of a start-and-stop conversation. 

These usability updates are especially important because for all its power, ChatGPT is still a chatbot and part of the challenge has been figuring out exactly how to talk to it, using language and syntax that will get the type of response the user wants. That's true of both the older GPT-3.5 version and the more recent GPT-4 version. 

Using voice chat in the ChatGPT mobile app, I asked the new ChatGPT-4o what it thought the most important new features were. It said the following, in a fairly natural-sounding spoken voice: 

"ChatGPT-4 Omni is designed to offer a more interactive and immersive experience by integrating visual, audio, and text inputs. This means it can analyze images, understand voice commands, and provide responses in a more engaging way. The goal is to make interactions more natural and versatile, allowing users to switch between different modes of communication seamlessly."

I tried another new feature and interrupted the AI to ask specifically what was different in GPT-4o versus GPT-4 or GPT 3.5, and the voice assistant answered: 

"ChatGPT-4 Omni stands out by incorporating multimodal capabilities, meaning it can handle and interpret images and voice inputs in addition to text. This is a significant upgrade from ChatGPT-4 and 3.5, which were primarily text-based. Omni also features enhanced context retention and understanding, making conversations more fluid and natural over extended interactions. These improvements aim to provide a more dynamic and interactive user experience compared to the purely text-based predecessors."

It certainly felt like a fluid conversation, and the response speed was at least as good if not better than my daily interactions with other voice assistants like Amazon's Alexa. But where I really felt an immediate difference was in how it responded to text input. 

Screenshot: Dan Ackerman  

I asked a reasonably technical question about why someone would want to utilize WSL, or Windows Subsystem for Linux, for use in data science or machine learning. The existing GPT-4 model answered in 41 seconds, while the new GPT-4o model gave a similar-length answer (both were around 300 words) in under 10 seconds. 

The best part is that users of the free version of ChatGPT will have access to the new GPT-4o and the existing GPT-4 tools. Previously, they were restricted to using the older GPT-3.5 model. Usage is limited for free users, but that's still a big upgrade for free users. ChatGPT Plus, Team, and Enterprise users will have higher usage limits. 

Keep in mind, however, this is still cloud-based AI (OpenAI says a desktop app is coming soon), so for full control over your AI tasks and better security for sensitive or proprietary data, a local AI solution might be the way to go. Expect to hear a lot more about CPU and GPU hardware for local AI acceleration in the coming months.