On December 6, Google unveiled Gemini 1.0, a multimodal large language model in three variations: Gemini Ultra, Gemini Pro and Gemini Nano. Many were quick to term it as an answer to ChatGPT as, on paper, Gemini performed much better than GPT. But does it provide a viable alternative to ChatGPT?
The answer to this question remains vague. Bard, Google’s AI assistant, is powered by Gemini Pro, an intermediate model, while Gemini Nano will be available for Pixel 8 Pro users.
Google released a YouTube video titled “Hands-on with Gemini: Interacting with multimodal AI”, showcasing the AI’s incredible performance. The video was impressive, with the presenter drawing a duck, playing guess-the-country, placing various objects, and interacting with Gemini.
However, numerous reports suggest that the interaction detailed in the video was not real-time and was a series of text prompts with still images.
This could have been less embarrassing for Google if the model had been seamless. Users quickly pointed out that Gemini was showing incorrect answers and even sometimes asked the users to Google it themselves.
Google DeepMind researchers compared Gemini with GPT-4, and as per the data, Gemini should perform well in most cases. In every capability, general, reasoning, math, and coding, Gemini scored better than GPT-4 except for “commonsense reasoning for everyday tasks,” where GPT-4 performed better.
Enthusiasts were quick to pitch the two LLM-based AI models. ArsTechnica tested both the AIs on various everyday tasks that a user with no computer background will do. GPT performed better in coding and generating summaries, while Gemini-powered Bard scored well in math. However, GPT appeared a clear winner.
On the other hand, Gemini Pro outperformed GPT3.5 in Tom’s Guide comparison. However, in this comparison, GPT could overpower Gemini Pro in coding.
Is Gemini the answer to GPT?
As of now, we can’t be certain. Recently, OpenAI debuted GPT-4 Turbo and custom GPTs, which are more cost-effective with an expanded context window. The company has also launched an experimental program to fine-tune GPT4. Custom GPTs are designed specifically for the task and could be useful for different technical professions.
Currently, Google seems to lag significantly behind Microsoft-backed OpenAI.
Gemini Pro-powered Bard has not performed much better than GPT in most comparisons and has underperformed in some areas.
“We really don’t know anything about Gemini Ultra. Does it beat GPT-4 for real? If so, why by such a small amount? Two options: 1) Gemini represents the best effort by Google, and the failure to crush GPT-4 shows the limits of LLMs approaching 2) Google’s goal was just to beat GPT-4,” tweeted Ethan Mollick, Professor at Wharton.
Google wants to be the winner of the AI race. With the lacklustre launching of Bard, Google decided to up its game with Gemini. However, the other side is not budging either. Microsoft has integrated Copilot into Windows 11 and plans to bring it to Windows 10. Moreover, with custom GPTs, many more will come into the GPT fold.
That being said, Gemini is still in its first phase and is performing better. The subsequent phases might result in the AI being more advanced.
In the News: Apple might discontinue some iPad models in 2024
