Google has introduced a new AI model called Gemini, which it claims outperforms ChatGPT in various tests and demonstrates advanced reasoning across multiple formats. Gemini is a multimodal model that can comprehend text, audio, images, video, and computer code simultaneously. It will be integrated into Google products, including the search engine. The model comes in three versions: Pro, Nano, and Ultra. The Ultra version, the most powerful iteration, will undergo external red team testing and will not be released publicly until early 2024. Google is in discussions with the UK government about testing Gemini through the AI Safety Institute. While Gemini shows promising capabilities, hallucinations or false answers remain an unresolved research problem. Google released videos showcasing Gemini’s abilities, such as understanding handwritten physics homework and identifying drawings. Concerns over AI range from disinformation to the development of superintelligent systems. Gemini represents a step towards artificial general intelligence (AGI), but there are still areas of research and innovation needed. Ultra, the most advanced version, outperformed human experts on a multitasking test and will power a code-writing tool called AlphaCode2. The data used to train Gemini was sourced from various places, including the open web, which has raised concerns from the publishing and creative industries.
