At the Google I/O event earlier this year, Sunar Pichai announced that development of Gemini's new LLM model had begun. According to a report by The Information, the launch was supposed to take place this week but has been postponed to early next year.
Among the issues faced with Gemini is that it has not been able to provide satisfactory answers to questions asked in non-English languages. To give competition to Llama 2 and GPT-4 Turbo, Gemini should have features that can already be AI'd by these two Google competitors.
According to the original announcement, Gemini is more powerful when compared to the currently used PaLM 2. Gemini was developed using the expertise of DeepMind which has already produced AI that can predict the weather more accurately and found 2.2 million crystal structures equivalent to 800 years of research.
The question for ordinary users is how Gemini will be integrated into Google products in the near future. With smartphones with AI available to be a hot trend next year, can Gemini be offered directly on devices with a smaller number of parameters?