- December 7, 2023
- AI Projects
On Wednesday, Google introduced its groundbreaking initiative, Project Gemini, a sophisticated artificial intelligence system designed to mimic human behavior, sparking further conversations around the profound implications and possible risks of such advanced technology. Google plans to deploy Gemini in stages, initially integrating simpler versions, dubbed “Nano” and “Pro,” into its conversational AI agent Bard and the upcoming Pixel 8 Pro smartphone.
Leveraging Gemini’s capabilities, Bard is set to become more perceptive and skilled at undertaking planning-related tasks. The Pixel 8 Pro will gain the ability to swiftly distill audio recordings on the device and craft automatic responses for popular instant messaging platforms, starting with WhatsApp, as per Google’s announcement.
The most notable breakthroughs of Gemini are expected to arrive next year with the release of the “Ultra” model, powering an enhanced chatbot called “Bard Advanced,” initially accessible to a select group of users for testing. Despite initial rollouts in English, Google executives assert that multi-language support will follow without difficulty.
Google’s demo of Gemini for the press suggested that “Bard Advanced” could achieve unparalleled AI multitasking, comprehensively processing and comprehending diverse content including text, images, and video all at once. In time, Google plans to embed Gemini within its predominant search platform, though the specifics of this integration remain undeclared.
Demis Hassabis, the CEO of Google DeepMind, the division responsible for developing Gemini, hailed the launch as a landmark achievement in AI and a new chapter for Google. Google secured DeepMind, based in London, after outbidding competitors such as Meta, and effectively fused it with their internal “Brain” team to concentrate on developing Gemini.
Google has highlighted that the technology showcases exceptional problem-solving abilities, particularly in mathematics and physics, nurturing hopes of potentially groundbreaking scientific advancements that could enhance human life. There is a faction concerned about AI outpacing human intellect, leading to job losses and potentially more harmful outcomes, such as the spread of false information or even activating nuclear arsenals.
In a blog post, Google CEO Sundar Pichai emphasized the company’s balanced yet ambitious approach in developing AI, prioritizing transformative research while instilling safeguards and cooperating with authorities and specialists to mitigate risks. Gemini’s debut is poised to elevate the rivalry within the AI sector, especially against OpenAI and its seasoned adversary, Microsoft.
In March, as Google unveiled Bard, OpenAI introduced GPT-4 and continued adding features to cater to both consumers and businesses. This expansion includes a recently added function that allows the bot to analyze images. OpenAI, along with other AI ventures, is also vying for market share against its backer, Microsoft, which enjoys exclusive rights to use OpenAI’s technology thanks to its substantial financial investment.
Microsoft’s collaboration with OpenAI has substantially enhanced its valuation, as investors anticipate AI becoming an immensely profitable sector. Similarly, Alphabet, Google’s parent company, has seen its market value increase significantly, although Alphabet shares experienced a slight dip after the announcement of Gemini. The tightening bond between Microsoft and OpenAI, coupled with the latter’s push towards product monetization, sparked apprehensions that it might be veering from its foundational aim of safeguarding humanity as AI evolves.
Eli Collins, the vice president of product at Google DeepMind, expressed his wonder at what Gemini could accomplish. At a recent virtual briefing, Google chose not to disclose the number of Gemini’s parameters, a key yet not the sole metric of a model’s sophistication. A research paper released the same day demonstrated Gemini outperforming GPT-4 on various assessments, including multiple-choice exams and elementary-level math, but it also recognized the ongoing challenge of elevating AI’s reasoning capabilities.
Some experts argue there are inherent limitations within large language models, which are configured to predict subsequent words in text sequences and can sometimes generate inaccurate fabrications called hallucinations. Collins acknowledged the progress made with Gemini in enhancing factual accuracy, naming it their premier model in this aspect, but acknowledged that solving the problem entirely remains an area of active research.
