Google made its next step in artificial intelligence on December 6 by unveiling Project Google Gemini. An AI model trained to act like humans will likely stir debate about its benefits and risks.
Google is rolling out Gemini in stages; their AI chatbot Bard and Pixel 8 Pro phones will utilize less sophisticated versions of Gemini.
Google promises that Bard will be more intuitive and excel in tasks requiring careful planning. At the same time, Gemini on the Pixel 8 Pro can quickly summarise recordings and send automatic responses via messaging services.
Gemini Is Available In India
Google recently integrated Gemini Pro with Bard, its direct competitor to ChatGPT. Bard powered by Gemini allows users to interact with it via text.
Google has pledged support for other modalities in the “short” future; currently, English-only updates are distributed throughout 170 countries, including India, allowing Bard users to take advantage of Gemini’s capabilities.
What is Google Gemini?
Google claims Gemini to be the world’s most advanced AI model. These claims can be verified using Bard, Google’s chatbot. Gemini Google was designed as a multimodal AI tool from its inception – one of its key advantages over other popular models – which Google demonstrated through multiple demonstrations showing its interaction with various media such as text, audio, and video to make interactions seem more human-like.
Gemini Capabilities
Google extensively tested its new artificial intelligence model Gemini across various tasks. Gemini Ultra’s most advanced version outshone human experts on 30 of 32 academic tests commonly used for AI research and development.
Furthermore, Gemini Ultra scored 90.0% in an MMLU (Massive Multitask Language Understanding) test, which assesses problem-solving as well as knowledge across 57 subjects such as maths, physics, and history – this marks the very first time an AI model outshone human experts on such an exam!
Coding Ability
Gemini’s first version can create and interpret high-quality code for popular programming languages like Python, Java C++, and Go. Google claims it to be one of the world’s most advanced artificial intelligence models for coding.
Gemini Ultra performed admirably across several coding tests using HumanEval as the industry standard and Natural Code utilising information generated by authors rather than information found online.
Gemini can also power more advanced coding programs. AlphaCode was one of the first AI-powered systems that could produce code competitive enough to compete in programming contests;
Google later created AlphaCode 2 using an enhanced Gemini version, which excels at solving mathematical and theoretical computer issues.
“Bard Advanced” in Early 2024
Gemini will experience its greatest advances by 2024 when the Ultra model is used to launch “Bard Advanced,” an upgraded chatbot initially only available to a select few.
Google executives explained at a press briefing that AI would initially only work in English; however, they assured that it would eventually support other languages.
Google’s Bard Advanced, known by its former moniker as Gemini, could achieve unparalleled Gemini AI multitasking by simultaneously understanding and recognizing presentations containing text, photos, and video.
The timing of this transition remains unclear. Demis Hassabis is the CEO of Google DeepMind, the artificial intelligence (AI) division responsible for developing Gemini.
Nearly ten years after acquiring DeepMind from Meta, the parent company of Facebook, Google integrated DeepMind with “Brain” to focus on Gemini development exclusively.
Triggering a debate
Google boasts of AI’s problem-solving abilities in math and science, leading many AI enthusiasts to anticipate scientific advancements that improve human lives due to this technological advancement.
An opposing side in the AI debate fears that technology will eventually surpass human intelligence, potentially leading to massive job losses and more destructive behaviour, such as spreading falsehoods or setting off nuclear bombs.
Sundar Pichai, Google’s CEO, wrote recently on his blog that the company was approaching their work with courage and responsibility.
This requires us to be ambitious when pursuing capabilities that benefit people and society while at the same time building safeguards into our approach and working collaboratively with experts and governments to mitigate risks.
Gemini’s introduction is expected to ratchet up competition between San Francisco startup OpenAI and longtime rival Microsoft in AI development, further intensifying competition.
OpenAI GPT-4 to Be Challenged by Gemini
OpenAI was deep into developing GPT-4 when they released their free ChatGPT tool at the end of 2017. This AI-powered chatbot became an international sensation and raised awareness of its commercial potential – prompting Google to remove Bard from their search results.
OpenAI released GPT-4 shortly after Bard debuted in March 2023 and since has added many new features for consumers and businesses alike, including one unveiled last November that enabled chatbots to analyze images.
OpenAI faces competition for business from other AI startups like Anthropic and Microsoft – which holds exclusive rights over OpenAI technology due to the billions invested by both parties.
Microsoft has experienced great success from their alliance, as evidenced by its market value increasing by more than 50% year-to-date, as investors believe AI will become a boon for tech sector investors.
Alphabet parent Google has also experienced great gains as its market value has skyrocketed more than $500 billion or 45 percent year-to-date;
Alphabet shares fell slightly on December 6 despite all of the excitement around Gemini recently.
Microsoft’s increasing involvement with OpenAI over the last year and OpenAI’s aggressive commercialization initiatives have raised concerns that OpenAI has drifted from its original mission of protecting humanity as technology progresses.
Fears were compounded in November 2023 when OpenAI’s Board suddenly fired CEO Sam Altman over unreported trust issues.
It sparked a widespread backlash that threatened to tear apart its operations, leading to the mass exodus of AI engineers to Microsoft.
After restructuring its board and reinstating Altman as CEO again after receiving threats of destruction by opponents of OpenAI, Microsoft took immediate steps against OpenAI as well.
OpenAI may find itself where it needs to prove that its technology outwits Google DeepMind’s technology. Eli Collins, Vice President of Products at Google DeepMind, spoke highly of Gemini and was “in awe” of its abilities.
Google declined to disclose the number of parameters in Gemini, which is one measure but not the only indicator of its complexity.
Instead, in a white paper published December 6, they described a version of Gemini that outperformed GPT-4 in multiple-choice tests, grade-school mathematics benchmarks, and other areas; they acknowledged, however, that AI models still struggle to reach higher levels of reasoning.
Computer scientists believe that large language models are limited in their ability to do anything beyond predicting the next word in a sentence, often making errors known as hallucinations.
Collins reported that Gemini has made great strides toward factuality; indeed, it may even be considered the best model. Still, it remains an open problem that needs solving.
What is Gemini’s Approach To Hallucinations, Safety, And Security?
Google announced it would add “new protections” to Gemini to consider its multimodal capabilities and evaluate potential risks at every stage of development, testing and mitigating them as necessary.
Furthermore, they claim to conduct “the most comprehensive safety assessments of any Google AI models (including bias and toxicity assessments),” with research into cyber offence and persuasion.
External experts and partners will help stress test their models on various issues to identify blindspots within Google’s internal evaluation process.
Is Gemini Better Than ChatGPT 4?
Gemini may be more flexible than ChatGPT4, although it’s hard to tell exactly. Gemini allows video playback on devices without the internet and gives it an edge; plus, its free price point makes Gemini an attractive alternative.
