Hurdles beyond the hype of GPT-4o’s impact on businesses

“If it writes like a human and talks like a human, it probably is a human,” is a twist on a popular phrase by Redditors.

But this may be hard to ascertain for many given the current capabilities of AI – especially after OpenAI launched its voice assistant product GPT-4o.

The “o” in the new upgrade of GPT-4 stands for “omni,” which signifies it can do everything – talk, see, and understand like a human.

This model’s ability to respond in a human-like manner is what many experts believe would have a major impact on many businesses.

“A full-stack voice and workflow solution, offered by GPT out of the box at a very low cost, will be highly attractive to developers,” Hemant Mohapatra, partner at Lightspeed India, tells Tech in Asia.

According to him, a lot of companies like Play.ht, ElevenLabs, and Hume are already in the market using the text-to-speech and speech-to-text approach, and they will feel pressure from developers who use GPT-4o.

For instance, in this video, the AI demonstrates a similar conversational skill to a human and even displays emotions by cracking a joke, besides giving suggestions as requested.

My initial reaction is that it bears a near resemblance to Jarvis, an AI assistant to Tony Stark in the Iron Man movie series. This fictional AI managed everything from Stark’s home and lab, operated the Iron Man suits, and acted as a personal assistant.

However, Jarvis is still far from reality, as GPT-4o is basically generating combinations of words that it thinks sound right, in a human tone.

And apparently, the AI’s not even good at it. Users on Reddit complained that the new model can’t follow instructions for a long time and that once it deviates from the instructions, it “basically becomes a lost cause and it’s easier just to start a new chat fresh.”

This shows how GPT-4o can’t take into account too many prompts while maintaining context – something that OpenAI will have to work on.

Beyond the hype

Though the use cases listed by OpenAI are promising, it’s still unclear how safe and practical they can be.

For one, the model now needs constant internet access and massive, localized data centers to process especially large amounts of video and audio information.

A Swiss Army knife of AI?

Not there yet