Google unveils Gemini, an AI model trained to mimic human behavior, to challenge ChatGPT

Google has introduced its new AI model called Gemini as a competitor to OpenAI’s ChatGPT. Gemini is designed to behave in human-like ways and improve reasoning abilities. The rollout of Gemini will occur in phases, with versions called “Nano” and “Pro” being incorporated into Google’s chatbot Bard and Pixel 8 Pro smartphone. Gemini will enhance Bard’s intuitiveness and planning capabilities and provide automatic replies on messaging services. The Ultra model of Gemini will be used to launch an advanced version of Bard in early 2023. Gemini will initially work only in English but will eventually diversify into other languages. Gemini demonstrated the ability to multitask by recognizing and understanding presentations involving text, photos, and videos. Google plans to integrate Gemini into its search engine as well.

CEO of Google DeepMind, Demis Hassabis, called Gemini a significant milestone in AI development and the start of a new era for Google. The technology’s problem-solving skills in math and physics are praised, raising hopes for scientific breakthroughs. However, concerns about AI surpassing human intelligence and resulting in job loss and harmful behavior are also present. Google aims to be ambitious in its AI research while building safeguards and collaborating with governments and experts to address risks.

Gemini’s introduction will heighten the competition between Google and other AI players. OpenAI, backed by Microsoft, released its AI model GPT-4, which gained fame and challenged Google’s Bard. OpenAI has been developing new capabilities and competing against other AI startups. Microsoft’s involvement with OpenAI has led to concerns that OpenAI has strayed from its original mission. Gemini’s launch may require OpenAI to prove the superiority of its technology compared to Google’s.

Google declined to share the exact parameter count of Gemini but released a white paper outlining its capabilities. Gemini outperformed GPT-4 on multiple-choice exams and grade-school math but still faces challenges in achieving higher-level reasoning skills. Some computer scientists see limitations in large language models like Gemini, as they can make errors known as hallucinations. Google aims to address these challenges through ongoing research.