By Christopher Lessler
On Feb. 15, Google announced Gemini 1.5, an update of the original Gemini 1.0 that was introduced on Dec. 6. According to Google’s blog, Gemini 1.5 has dramatically improved performance and outperforms Gemini 1.0 on 87 percent of the benchmarks used in Google’s model testing.
Gemini is based on Google’s previous Bard AI chatbot, which was originally released in March 2023 to a relatively lackluster reception. A few weeks later, Google announced it was merging Google Brain and DeepMind, its two main AI labs, to develop more AI-powered technologies. Google has now also folded its Duet AI tool into Gemini, meaning that from now on Gemini will be Google’s near sole focus as far as AI goes.
When Google originally announced Gemini 1.0, the company created a video showcasing many of its abilities, including item recognition, translating words, playing a simple guess-the-country game, explaining a sleight-of-hand magic trick, and offering creative suggestions for the use of two colors of yarn. Google’s video garnered backlash for not being captured in real time despite seeming to be, but instead cherry-picking the most successful responses that Gemini offered.
Google says Gemini outperformed ChatGPT in 30 of 32 key benchmarks, including in reasoning and image processing, although its margins of improvement tended to be minuscule. Gemini’s current and most powerful version, Ultra, also became the first AI model — according to Google — to outperform humans on MMLU, an interdisciplinary test covering 57 subjects, such as physics and law. ChatGPT might finally have some serious competition in Google’s Gemini AI language model.
Not long after ChatGPT’s launch, Google brought its co-founders Larry Page and Sergey Brin back to develop an AI service, after they had both retired in 2019. This development process is ongoing and has already resulted in Bard and its improved version, Gemini. Upon its launch, Gemini was not available in the E.U., and although the web version is now available, there is no app version in the E.U. It was speculated that Google was waiting on the E.U. Parliament to vote on its AI Act, which was unanimously passed on Feb. 13. Nonetheless, AI integration into smartphones is moving forward, with Google and Samsung having partnered to bring Gemini-powered AI tools to the Samsung Galaxy S24, S24+, and S24 Ultra phones. Some Gemini-powered features also exist on Google’s Pixel phones. Gemini has a free version called Gemini Pro, but a paid option called Gemini Advanced is also available via a $20 monthly subscription fee for the Google One AI Premium plan, which also comes with other Google-related benefits. The paid version can have more involved conversations and do more complicated tasks, including coding.
On Feb. 8, Google began asking Gemini users not to tell anything confidential to the AI, saying that human reviewers were seeing some select data to improve Gemini. To avoid this, users can also choose not to save Gemini conversations to their Google account. This comes after Google published a list of moral principles in 2018 as it anticipated developing the kinds of AI tools seen today. The list includes ensuring AI is safe, societally beneficial, and not unfairly biased. Google also committed to not pursuing AI applications for weapons, surveillance that violates “internationally accepted norms,” any harmful applications, or any applications violating international law.
Leave a Reply