A YouTuber shows that GPT-4 can do what Gemini had to fake

Google, not being able to launch an AI model without some sort of drama, had to admit to some shady behavior. The company admitted to faking some elements of its Gemini hands-on video, and it’s gotten some flack for it. Well, Google just can’t catch a break, as a YouTuber showed that GPT-4 can do some of what the search giant had to fake.

The video that Google posted showed a person interacting with Gemini through real-time conversations. The company admitted to Bloomberg that the conversation we heard in the video never happened. The Google team actually used stills from the video and text prompts for the results.

One of Gemini’s co-leads, Oriol Vinyals, said that Gemini can do everything in the video, and that’s technically true. However, the video was rather misleading.

GPT-4 did what Gemini had to fake

So, Google’s gotten its slap on the wrist, but a YouTuber came by to deliver a slap to the face of Google. A newcomer to the YouTuber scene, Greg Technology, posted a video where he did a few of the same things that we saw in the Google video. The Greg Technology video was much narrower in scope in terms of the input that he presented. For instance, we didn’t see him ask GPT-4 to generate images with balls of yarn or create music based on a crude drawing.

Rather, he asked GPT-4 to identify a hand gesture, identify Rock-Paper-Scissors, and identify a picture of a duck. It did a good job with the exception of a slip-up at the beginning. Another thing to note is that GPT-4 took much longer to reply than what we saw in Google’s video. It shows that the technology still has a ways to go before it can achieve what we saw in the original Google showcase.

We’re, undoubtedly, going to see Google come out with something to resemble what we saw in the video. However, that day is not here.