Google defended the demo, with DeepMind's VP of Research, Oriol Vinyals, stating that all user prompts and outputs in the video were real, albeit shortened for brevity. However, the company has been criticized for not allowing journalists and developers to experience the product firsthand. Critics argue that Google's approach could be misleading and that it should let people test Gemini in a public beta to demonstrate its true capabilities.
Key takeaways:
- Google has been accused of misrepresenting the capabilities of its new AI model, Gemini, in a demo video. The video showed Gemini responding in real-time to spoken prompts and image recognition tasks, but a disclaimer revealed that latency had been reduced and outputs shortened for the demo.
- Bloomberg columnist Parmy Olson criticized Google for this, arguing that it suggested a level of real-time interaction that wasn't accurate. Google admitted that the demo used still image frames from raw footage and text prompts, rather than real-time spoken prompts.
- Google has a history of questionable demo videos, which has led to increased scrutiny and skepticism. Olson suggests that Google is "showboating" to distract from the fact that Gemini lags behind OpenAI's GPT.
- Google defended the demo, with DeepMind's vice president of research and deep learning lead, Oriol Vinyals, stating that all user prompts and outputs in the video were real, though shortened for brevity. He argued that the video was made to inspire developers and show what user experiences with Gemini could look like.