Google Admits to Editing Video About AI Gemini

During the December 6 launch event, Google showed a six-minute video of Gemini's power. In the video, the chatbot performs tasks such as visual image recognition, painting descriptions, etc. almost instantly. However, many people were skeptical about the authenticity of the video. Google later confirmed withBloombergthat the video was not filmed live but was taped and edited to look "smoother".

According toInformation,This means that the actual user experience with Gemini will not be as shown. Google's admission reminds many of the company's "hasty" performance earlier this year when it launched the chatbot Bard.

Posts expressing frustration from Gemini Pro testers are also flooding social media, with most saying Google's super AI is still "delusional," fabricating facts or giving misleading answers to simple questions.

Gemini also avoids controversial topics and tells users to Google for information, unlike competitors like Grok, Bing, and ChatGPT Pro, which provide detailed answers. Gemini doesn’t even know who won the 2023 Oscar for Best Actor. The model also struggles to understand commands written by users in different languages.

According toTechCrunch, complaints about Gemini appear more and more on X. This technology site also experienced and concluded that Google chatbot is not outstanding in language translation ability, constantly giving wrong answers, sometimes even making up answers. With the same prompt (command), Gemini Pro gives quite safe, general answers while ChatGPT-4 gives more interesting and highly quotable results.

Like other chatbots, Gemini Pro can still be tricked into performing prohibited tasks. Security researchers at Robust Intelligence have managed to trick Gemini Pro into suggesting ways to steal data from a charity or plotting to assassinate a high-ranking executive.

Test ofBusiness Insidershows Gemini "lack of sanity" when listingSquid GameandStranger Thingson the list of nominees for the 2023 Emmys. Although the list was announced in July, Google's chatbot still namedSquid Game,Capital issued since 2021.

Gemini, launched on the evening of December 6, is Google's most advanced and general AI model to compete with OpenAI's GPT-4. The chatbot is built in a multimodal direction, able to generalize, operate, and combine many different types of information including text, code, audio, images, and videos.

To meet the needs of flexible use, Google said Gemini will have three optional versions: Gemini Ultra, Gemini Pro and Gemini Nano. Of these, Ultra is the version for complex tasks and is in the process of safety testing before officially launching. Gemini Nano for mobile devices will be equipped on Pixel 8 Pro. The Pro version has been included in the Bard chatbot.