Image of apps on a screen, including ChatGPT

istock.com/Robert Way

ChatGPT’s Latest Model GPT-4o Is More Human-Like

May 14, 2024

On Monday, the CTO of OpenAI Mira Murati showcased ChatGPT’s human-like spring update with a range of live demos. The latest version of the AI chatbot, utilizing OpenAI’s latest flagship AI model GPT-4o, has the ability to reason across audio, vision, and text seamlessly.

It appears the CEO of OpenAI, Sam Altman, wasn’t joking when he said the new update “feels like magic.” This recent upgrade has left rivals under pressure.

To begin with, ChatGPT’s voice and conversational capacity has upskilled significantly with GPT-4. This upgrade allows the chatbot to convey emotions using a range of different tones effectively.


During the demo, the new AI showcased a voice resembling that of an American female, similar to Scarlett Johansson’s in the movie “Her” by Spike Jonze. But OpenAI researchers also revealed a switch to a robotic voice at one point. According to an OpenAI spokesperson, initially, the audio output will be restricted to a few preset voices upon launch.

Not only did the voice resemble a human’s, but it also showed a talent for mimicking human speech patterns. The updated ChatGPT can laugh, inject humor, and adjust its tone in line with the input it receives.

The latest upgrade also includes the capability to understand certain “human cues.” For example, during a demonstration where a researcher was hyperventilating when practicing deep breathing, the chatbot responded humorously with, “Mark, you’re not a vacuum cleaner.”


Users can intervene with the chatbot, making conversations feel more fluid. There’s no need to wait for the AI to complete its response before asking a follow-up question or changing the subject.

According to an OpenAI representative, the chatbot can react to audio inputs at a rate comparable to humans, averaging 320 milliseconds per response.

After the event, Altman shared on X (previously Twitter) the movie title that a lot of people were reminded of after witnessing the demos.

The chatbot showed off enhanced skills in graph analysis, coding assistance, emotion interpretation, and acting as a math tutor by analyzing videos or images taken with a phone’s camera.

Meanwhile, throughout the whole demo, the voice assistant kept a friendly and upbeat composure.

In a separate online demonstration, GPT-4o analyzed video footage of the user’s surroundings, spotting the individual wearing an OpenAI hoodie and being surrounded by recording equipment. From this, the AI made out that the person might be preparing an announcement about OpenAI.

Despite encountering some issues, such as misinterpreting an image prompt or prematurely responding before the question was fully asked, these moments added a human-like touch to the chatbot.

During one instance, ChatGPT started responding to a prompt before the researcher displayed the equation on camera. The researcher then had to intervene and stop the chatbot mid-response. In response, the chatbot said, “Whoops, I got too excited. I’m ready when you are.”

Recent News