Big fail by GPT5.4- with a roast from Gemini at the end.
OpenAI's live GPT-4o demo stumbled on a simple task, prompting a direct, mocking response from its rival.
OpenAI's launch event for its new flagship model, GPT-4o, intended to showcase its advanced multimodal capabilities, hit an unexpected snag during a live demonstration. The model was presented with a visual reasoning task: analyzing a grid of nine squares where one was uniquely shaded. When asked to identify the pattern, GPT-4o failed to provide a correct or coherent analysis, stumbling publicly on what should be a straightforward problem for a state-of-the-art vision model. The awkward moment was captured and quickly spread across social media, turning a planned showcase into a viral example of AI fallibility.
Seizing the opportunity, Google's Gemini team engaged in a direct and public competitive roast. They posted a screenshot on social platform X showing their Gemini model successfully solving the exact same grid problem that stumped GPT-4o. The post was captioned with a sarcastic remark, essentially implying the task was trivial. This exchange underscores the fiercely competitive nature of the current AI landscape, where launches are live-tested and rivals are quick to capitalize on any misstep. It transforms product marketing into a real-time, public battleground for capability and reliability.
- OpenAI's GPT-4o failed a basic visual pattern recognition task during its public live demo.
- Google's Gemini team responded by posting proof their model solved the identical problem, with a mocking caption.
- The incident highlights how AI product launches are now subject to immediate public testing and competitive social media warfare.
Why It Matters
For professionals, it demonstrates that even top-tier AI models have brittle moments, and vendor competition is intensifying publicly.