Google’s new AI language model, Gemini, has taken the spotlight with its claims to outperform OpenAI’s GPT-4 in most tests. However, Microsoft has fired back, asserting that GPT-4 holds the upper hand when provided with the right prompts.
Gemini, now available in three versions – Nano, Pro, and Ultra – has generated excitement with its impressive capabilities. In 30 out of 32 commonly used tests, Gemini Ultra surpassed GPT-4, exhibiting superior performance across reading comprehension, math questions, Python coding, and image analysis. Though the differences between the models varied, with some tests only presenting marginal disparities, others demonstrated gaps as large as ten percentage points.
Gemini Ultra’s groundbreaking accomplishment lies in its triumph over human experts in massive multitask language understanding (MMLU) tests. It scored an impressive 90.0 percent, slightly surpassing a human expert’s score of 89.8 percent. Gemini’s triumph in such a diverse range of fields, including math, physics, medicine, law, and ethics, has fueled optimism for its potential applications.
While Google’s Gemini has sparked great interest, its roll-out will be gradual. Gemini Pro is now accessible to the public, integrated into Google’s chatbot Bard, while Gemini Nano is embedded in various functions on the Pixel 8 Pro smartphone. On the other hand, Gemini Ultra, still undergoing security testing, is currently only available to a limited number of developers, partners, and AI liability and security experts. Google plans to make it available to the public via Bard Advanced early next year.
However, Microsoft is not prepared to let Google’s claims go unchallenged. Microsoft researchers recently published their research on Medprompt, an innovative approach involving modified prompts or inputs that led to improved results in GPT-4. By leveraging Medprompt, Microsoft’s GPT-4 excelled in numerous tests, including the MMLU test, where it achieved a score of 90.10 percent. The battle for AI supremacy remains fierce, and it remains to be seen which language model will ultimately reign supreme.
As Gemini and GPT-4 continue to vie for dominance, the future of AI hangs in the balance. With each model pushing the boundaries of language understanding and processing, the potential for groundbreaking advancements is within reach. The journey towards the AI throne is far from over, and the developments in this fierce competition will surely shape the landscape of AI technology moving forward.
