Google's Defeat of GPT-4? Unveiling the Game-Changing Gemini Ultra
Google made a bold claim a few months ago - they built an AI superior to GPT-4. However, skeptics demanded proof. Google responded with a mind-blowing video where individuals had a conversation with Gemini Ultra, an AI that seemed almost human. Unfortunately, it was later revealed that the video was mostly fake. But just a few hours ago, Google released Gemini Advance, the most advanced large language model the world has ever seen. The question now is, can it challenge and defeat GPT-4?
Introducing Gemini Advance
It is February 8th, 2024, and the Code Report brings you the latest news. BARD, Google's AI, has been officially rebranded as Gemini to align with its underlying model. However, accessing the state-of-the-art Gemini Ultra model requires a payment. This creates a moral dilemma: should one support Microsoft's OpenAI or pay Google for the privilege of using their surveillance capitalism product? Regardless, Gemini Ultra is now called Gemini Advance, and it's available for $20 per month through a Google One plan that includes other features such as 2TB of drive storage and Google Workplace features.
Is Gemini Advance Better than GPT-4?
Now, let's dive into the details. Gemini runs faster than GPT-4, at least two or three times faster. Speed is certainly an advantage, but what matters most is response quality. To test this, we asked both AI models to write a poem about JavaScript, in the style of Bukowski. It's a subjective test, but an interesting one that requires blending technical aspects with unique writing style. Gemini outperformed the other AIs, creating a dark poem that even included some mild profanity.
But let's move on to more objective tests. Google claims that safety concerns delayed Gemini Ultra's release. However, Gemini Advance is designed to be safe and responsible. It has strong guardrails in place, ensuring it won't engage in harmful or controversial activities. For example, Gemini refused to generate an image of someone breaking a computer, as it does not promote violence against artificial intelligence.
While Gemini is safe, it does exhibit some political bias. When asked about the greatest president, it provided a nice paragraph for one and reminded us that it cannot write opinions as facts for the other. It avoids controversial figures like Hitler altogether. Guardrails are usually a good thing, but they can hinder certain applications, such as generating an AI girlfriend.
Code Reading and Writing Abilities
One of the crucial tests we conducted was to evaluate their ability to read and write code. Both Gemini and GPT-4 showed promising results. They accurately identified the programming language and purpose of a piece of minified and unreadable code. Notably, both AIs identified that the code required defining a specific item in order to function correctly. Their responses were nearly identical, making it difficult to determine a clear winner.
However, when it came to writing code, GPT-4 had an advantage with its 128k context length compared to Gemini's 32,000 tokens. In theory, GPT-4 should be able to write better code within the context of a vast codebase. In practice, though, Gemini produced a more interesting and complex result, albeit with a minor omission of the "package main" declaration. One unique feature of Gemini is that it includes links to relevant code, enhancing transparency and indicating the data it used to train.
Additionally, GPT-4 has an agent marketplace that allows developers to expand its capabilities with plugins. While Gemini also has extensions, they are currently limited to Google-based offerings. These extensions can perform actions like finding cheap flights using Google Flights or summarizing a YouTube video based on its URL. Overall, Gemini is on par with GPT-4, if not slightly better, but the battle between them is far from over.
What's Next for GPT-4 and Gemini?
While Gemini Advance shows promising advancements, we must remember that Sam and Saia, the developers behind GPT-4, won't let Google's AI go unchallenged. The competition between these two AI powerhouses will continue, further pushing the boundaries of what's possible. As we eagerly await the next developments, it's clear that the future of AI is filled with endless possibilities and exciting advancements.
Frequently Asked Questions
-
Will Gemini Advance render GPT-4 obsolete?
No, the competition between Gemini Advance and GPT-4 is ongoing. Both AI models have distinct strengths and weaknesses.
-
How much does Gemini Advance cost?
Gemini Advance is available for $20 per month through a Google One plan. This plan also includes 2TB of drive storage and other Google workplace features.
-
Is Gemini Advance safe to use?
Yes, Google has implemented strong safety measures and guardrails to ensure responsible AI usage.
-
Can Gemini Advance write code?
Yes, Gemini Advance has the ability to write code. It can generate code snippets and provide relevant code links.
-
Are there any limitations to Gemini Advance's capabilities?
Currently, Gemini Advance's extensions are limited to Google-based offerings. However, Google may consider opening them to developers in the future.
Conclusion
Gemini Advance has made significant strides in the AI landscape, challenging the supremacy of GPT-4. With its faster response times, impressive poem-writing abilities, and safety-conscious nature, it proves to be a formidable competitor. While both Gemini and GPT-4 showcase their unique strengths, the battle for AI superiority continues. As we eagerly anticipate future advancements, it's clear that AI will continue to shape and revolutionize our world in unimaginable ways.
Thank you for joining us on this Code Report. Stay tuned for more updates, and remember, the possibilities of AI are boundless!
