Last week Google showed off its new AI

The initially convincing impression of the Gemini AI turns out to have been edited. (Source: Google via YouTube)

On Wednesday evening, Google proudly presented the company’s new AI model: With “Gemini” they are launching an attack on ChatGPT & Co.

The sensational video on YouTube – with more than one and a half million views in the first 24 hours – shows, among other things, how the Gemini AI interacts with the user. A blue rubber duck was chosen as an example to show a natural dialogue.

However, this very conversation was heavily edited and does not exist in this form. A company spokesman confirmed this to the Bloomberg portal.

Link to YouTube content

The Gemini concept isn’t just a literal duck

As Google explains in the accompanying YouTube description, “for the purposes of this demo, the latency was reduced and the Gemini output was shortened for clarity.”

The extent of the editing can be seen in a post by Gemini co-development lead Oriol Vinyals on X (formerly Twitter). According to this, the user input and output are real, but the actual interaction with Gemini is classically based on uploaded images and text prompts.

Link to Twitter content

In short: The realistic “dialogue” with Gemini does not exist in this form. The YouTube presentation itself gave the impression that the AI ​​is talking to the test person and can recognize live images almost in real time.

As Vinyals further explains, the video is only intended to “illustrate what the user experience with Gemini might look like.” The clip is intended less to represent reality and more to “inspire developers.”

In the replies to the tweet, Google’s admission was sharply criticized. Factually incorrect content would not inspire developers, but rather mislead and raise false hopes.

The announcement of Gemini is just as bumpy as the launch of Google Bard: At the start, the text AI, which is a direct competitor to ChatGPT, was neither available in Germany nor was it able to convince in our long-term test:

Now the question is: In your opinion, did Google embellish the Gemini idea too much? Will this dampen your expectations of the new AI? Do you understand Google’s approach? Or do you not expect much from the new artificial intelligence anyway? Please let us know in the comments!