Google has released Gemini, its most powerful AI model, but has been accused of exaggerating its performance in a video. A column in Bloomberg said that Google distorted the capabilities of Gemini in a video. The video showcased Gemini's multimodal abilities, such as combining voice prompts and image recognition. Gemini appeared to quickly identify images and track a crumpled paper in a cup and ball game in real-time. However, Google included a disclaimer in the video description: "For demonstration purposes, we reduced latency and shortened Gemini's output." Columnist Parmy Olson was dissatisfied with this disclaimer. In her article for Bloomberg, she stated that Google admitted the video demonstration was not done in real-time with voice prompts, but instead used static image frames from the original video and wrote text prompts for Gemini to respond to. "This is very different from what Google seems to be implying: that a person can have a smooth voice conversation with Gemini, which can observe and respond to the world in real-time," Olson wrote. Google often edits demonstration videos, especially to avoid technical glitches during live presentations. In this case, Olson said Google was "showing off" and misleading people to overlook the fact that Gemini still lags behind OpenAI's GPT. Google disagrees. It cited an article by Oriol Vinyals, the Vice President of Research and Head of Deep Learning at Google DeepMind (also co-lead of Gemini), to explain how the team created the video. "All user prompts and outputs in the video are real and shortened for brevity," Vinyals said. "The video showcases what a multimodal user experience built with Gemini could look like. We made it to inspire developers." He added that the team provided Gemini with images and text and asked it to respond by predicting what would happen next. While this is one way to handle the situation, it may not be the right approach for Google, as it has already been caught off guard by OpenAI's significant success this year. If Google wants to inspire developers, it should do so not through carefully edited flashy videos that can be seen as distorting the capabilities of AI, but by allowing journalists and developers to truly experience the product.