• This week Google officially launched its latest foundational model for generative AI – Gemini.
• Available in three formats based on user need, Gemini is not yielding the results that many hoped.
• Google also confirmed that it staged a video demo of Gemini AI, bringing into question its actual readiness.

Google this week officially launched its latest generative AI foundation model, Gemini, with it being tipped to be a worthy competitor to OpenAI’s ChatGPT and in particular GPT-4.

The first 24 hours of its release, however, have not gone as smoothly or promisingly as Google may have hoped. The first issue, as Engadget recently reported on, pertains to the hands-on demo video (embedded below) of Gemini AI.

While the video is impressive, it turns out that the video itself was staged, per an opinion piece by Bloomberg (paywall), as elements were sped up to make it look like results were generated faster than they actually were in real-time.

The video description declares as much, but the voice interaction within the video, did not actually happen either, which brings into question just how legitimate on a hands-on demo it is.

This adds credence to claims a few days ago that Google was reportedly preparing to delay the release of Gemini into early next year, with its accuracy and response to non-English prompts flagged as a potential issue.

Contrary to the reports though, Google has pushed through with its release in order to keep to the before the end of 2023 timeframe it had imposed on itself when the development of this foundation model was announced at Google I/O earlier this year.

If early user feedback is anything to go by, a little more time to train or incubate may have been the better play, as some have reported less than satisfactory performance from Gemini, which as mentioned is billed as a potential rival to what OpenAI is making available.

To that end, TechCrunch spotted some frustrated posts from users of Gemini Pro on Bard. Here the AI does not seem to have access to up-to-date information, with it unable to list the 2023 Oscar winners and in particular not knowing whether Brendan Fraser took home the Best Actor award earlier this year (which he did).

Instead of mentioning Fraser, the AI said Brendan Gleeson won Best Actor, with the confusion possibly coming in as the actor was nominated for his Supporting Role in the Banshees of Inisherin earlier this year.

Much like Microsoft’s AI-imbued Bing, which too struggled with more recent queries, it may simply be a case of the data that Gemini is working with simply not being up-to-date. Given that Google is the world’s repository for all information online though, it is a rather concerning error.

There are more examples of Gemini not performing as expected, such as delivering incorrect code when prompted or not being able to handle translation tasks effectively.

Either way, with Google promising constant improvements to the performance of Gemini over time, it needs to push out said updates/enhancements rather quickly, or its late 2023 release would have been for naught.