Google Gemini, generative artificial intelligence (AI) model and chatbot created by the search engine company Google, which uses large language models (LLMs) to “converse” with users and generate content.
Bard (Google Gemini’s original name) was announced by Google CEO Sundar Pichai in February 2023 as an “experimental conversational AI service, powered by LaMDA.” LaMDA (Language Model for Dialogue Applications) is Google’s conversational program. Bard was released a month later. In June 2022 Google engineer Blake Lemoine claimed that LaMDA was sentient, which the company denied. (Lemoine was placed on administrative leave after going public with his claims and then fired that July).
Bard received criticism as soon as it was announced, as Google’s promotional demo showed an unintentional error, providing incorrect facts in a response to the prompt, “What new discoveries from the James Webb Space Telescope can I tell my 9 year old about?” Following the error, Google’s parent company Alphabet’s market value slid by $100 billion. Shortly after Bard’s launch, Bloomberg News released a report featuring screenshots of internal messages from Google employees. One employee said in an internal message, “Bard is worse than useless: please do not launch.” Another employee reported that Bard was a “pathological liar.” The testimonies suggested that Google had rushed Bard’s launch, bypassing ethics-based tests, in order to keep up with competitors such as Microsoft and OpenAI. In April 2023 Google consolidated two of its AI research groups, Brain and DeepMind, into Google DeepMind, a move that Pichai claimed would “significantly accelerate” Google’s progress in AI.
In December 2023 Google released an updated version of Bard that used a new LLM, named Gemini, claiming that the technology outperformed OpenAI’s GPT-4. Google co-founder Sergey Brin notably returned to Google to assist in creating the new system.
Google initially divided Gemini technology into three categories. Nano powers features on some of Google’s Android devices and can help make the devices more accessible. For example, the Google Pixel 8 Pro includes the TalkBack feature, which helps those with low vision interact with their device using touch and speech. Pro was added to Bard shortly after Gemini was announced and was marketed as having strong performance across a variety of tasks, such as summarizing reports and generating computer code. Ultra, which launched in February 2024, is claimed to be the fastest and most high-quality model. In addition to text, Gemini is also trained on images and sounds, making it multimodal, or capable of combining multiple types of information, such as text and images. For example, given an image, Gemini can describe the image and alter it. A few months after the launches of the initial three models, Google released Gemini 1.5 Pro, which it claimed was faster-performing. To address user concerns regarding the bulk of the software, Google then released Gemini 1.5 Flash, which it claimed was a lighter weight than its predecessor.
In early 2024 Google renamed Bard to Gemini. When the platform first launched, it was available only in English but expanded to include Japanese and Korean the following week. As of April 2024 the Gemini web app is available in more than 40 languages. Google also launched Gemini Advanced, claiming, “Gemini Advanced is far more capable at highly complex tasks like coding, logical reasoning, following nuanced instructions, and creative collaboration.”
The month of its launch, Gemini gained negative attention when social media posts emerged showing that the platform created inaccurate (and sometimes offensive) images that assigned incorrect genders and races to certain groups of people. Commentators, such as Tesla CEOElon Musk, criticized the algorithm, which engineers had misguidedly optimized to try to avoid bias favoring white males but which had veered too far in the opposite direction. For example, the model generated an image of a Nazi-era German soldier as an Asian woman. Following the reports, Google paused the platform’s ability to generate images of humans. Brin acknowledged that the company had “definitely messed up on the image generation,” stating his belief that “it was mostly due to just not thorough testing and…definitely, for good reason, upset a lot of people.”