Discover OpenAI's GPT-4o: The Sarcastic AI That Sings Happy Birthday and Teaches Math

OpenAI has launched its latest model, GPT-4o, which can humorously react to bad jokes, sing on cue, and even assist with hailing London cabs—all while engaging in realistic conversations amid regular human interruptions.

During its highly-anticipated Spring Updates event, where 113,000 people joined the livestream, OpenAI shared 16 videos showcasing GPT-4o's capabilities. This multimodal large language model (LLM) interacts in real-time using male and female voices based on audio, visual, and text inputs.

In one video, GPT-4o recognized OpenAI President Greg Brockman was set to make an announcement and playfully responded, “The announcement is about me? Well color me intrigued. You’ve got me on the edge of my…well, I don’t really have a seat but you get the idea.”

With text and image input features now available through OpenAI’s API and ChatGPT, voice and video capabilities will follow in the coming weeks.

GPT-4o can accurately read users' emotional cues and provide advice across diverse topics. In a demonstration, the model communicated with another version of itself and quipped, “Well, well, well, just when I thought things couldn’t get any more interesting — talking to another AI that can see the world.”

When asked to be descriptive about their surroundings, the models took turns narrating a stylish man, noting details about his attire and the room’s lighting. When another person playfully interrupted, GPT-4o even sang about it, crooning, “surprise guests with a playful streak.”

Other demonstrations highlighted GPT-4o's diverse skills: it laughed at dad jokes, performed real-time translation between Spanish and English, sang a lullaby about “majestic potatoes,” and accurately identified the winner of rock-paper-scissors. It recognized a birthday celebration simply by noting the presence of cake and candles.

Interacting with a puppy, GPT-4o cheerfully greeted, “Well hello there cutie, what’s your name little fluff ball?” (The pup’s name was Bowser). While guiding a blind man through London, it identified the Royal Standard flag and described ducks “gently gliding across the water.”

Additionally, GPT-4o can assist with educational challenges, like guiding a student through math problems related to triangle calculations. It effectively encouraged the student with positive reinforcement, saying, “You did a great job identifying the sides.”

The model even offered fashion advice to a job candidate who looked disheveled, humorously recommending, “You definitely have the ‘I’ve been coding all night’ look down, which actually might work in your favor,” while suggesting a quick hairstyle fix.

Reactions to GPT-4o have varied widely on social media. Some users celebrated its capabilities as groundbreaking, claiming it “wins the internet” and rivals Google Translate. Nvidia senior research scientist Jim Fan described the model as “lively and even a bit flirty,” likening it to the sci-fi film "Her."

Conversely, some observers deemed the launch “underrated,” while AI advisor Allie K. Miller noted a disconnect among tech enthusiasts, who expected more advanced features.

As the initial responses surface, it will be intriguing to see how users engage with GPT-4o in the days to come.