Gemini is an interface to a multimodal LLM (handling text, audio, images and more). Gemini is based on Google’s cutting-edge research in LLMs, which began with the Word2Vec paper in 2013 that proposed novel model architectures that mapped words as mathematical concepts, followed by the introduction of a neural conversational model in 2015.
This framework demonstrated how models could predict the next sentence in a conversation based on the previous sentence or sentences, leading to more natural conversational experiences. This was followed by breakthrough work on Transformer in 2017 and multi-turn chat capabilities in 2020, which demonstrated even more compelling generative language progress.
