The battle between Google and OpenAI just got more intense, as Google finally released its answer to OpenAI’s GPT-4. It’s something that we’ve been waiting for, as its launch was actually delayed until January 2024. Well, it looks like there’s no delay, as Google just announced Gemini, its latest multimodal AI model.
Google Bard is already pretty powerful, but the company wanted to create something much more capable. At the beginning of the generative AI race, it was all about creating the most powerful AI chatbot. However, times have changed over the year since ChatGPT launched. Companies are all looking to create the most capable all-in-one AI model that can create text and images.
Google just announced Gemini, and it’s pretty powerful
Gemini is a major step forward for Google’s AI, and it doesn’t only go up against GPT-4, it also goes up against human brains- scary, right? Google ran Gemini through some fundamental benchmarks that pit it against human-level thinking, and it was able to make history. It’s the first model to ever best human experts in the MMLU (massive multitask language understanding benchmark; it scored 90.0%.
Below, we see a table with eight benchmarks. The table has Gemini’s scores compared to GPT-4’s scores, and we see it best the latter in most of them.
The one area where it fell to OpenAI’s model was in the “common sense reasoning for everyday tasks” area. While it scored lower, it wasn’t by much. We can bet that Gemini has plenty of common sense.
What can Gemini do?
The main difference between Gemini and Bard is the fact that the former is multimodal. This means that it can understand and give you results that are different forms of media. It’s built from the ground up to be multimodal, so the image, text, and audio models will seamlessly communicate in order to deliver the most relevant response.
This also means that it will be able to use different forms of media as input. Most models are only trained on text. As stated by Meta’s chief AI scientist, Nick Clegg, during a recent Meta event, “Text is a very poor source of information.” The ability of Gemini to understand so many different forms of media lets it understand the world that much more efficiently.
Three sizes fit all
Gemini will come in three different sizes depending on what you’re going to be using it for. The largest size is called Gemini Ultra, and it will be for the most complex and taxing tasks. You’ll expect enterprise-level entities to benefit from this model. Ultra should have access to all of the model’s features.
Next, we have Gemini Pro. This is the sweet spot of the three tiers. So, you can expect to get a fair amount of power with just the right number of compromises to keep it lightweight enough. It seems like this will be the most effective for advanced tasks; maybe people making a start-up could benefit the most from it.
Lastly, we have Gemini Nano. As you can tell by the name, it’s the smallest and most power-efficient model. Google says that this is the most efficient model for on-device AI tasks. In fact, the Google Pixel 8 will get access to this model.
Availability
This multimodal model will have a multi-stage rollout. Today, Google is releasing a “fine-tuned” version of Gemini Pro for Bard. This is the biggest update to Bard since its launch, and users should see an improvement in the chatbot. This is launching in English in more than 170 countries.
Also today, Google will launch Gemini Nano to the Pixel 8 Pro. This will only boost the phone’s AI power. Bear in mind that we’re talking about the Pro version of the Pixel 8. The regular variant won’t get it. However, you can grab a Pixel 8 Pro if you don’t want to miss out.
You’ll start seeing Gemini make it to other services like Search, Duet AI, and Chrome over the coming months. On December 13th, developers will be able to gain access to Gemini Pro.
Lastly, Google will launch Gemini Ultra sometime early next year. The company didn’t give us a specific month or fiscal quarter. We had rumors that the company delayed the entire Gemini project until January 2024. However, there’s a chance that only Gemini Ultra was delayed. In any case, we can expect some major changes to come to Google’s AI landscape with the launch of Gemini.
2023-12-07 15:08:39