What you need to know
- Despite reports of it being delayed, Google Gemini has been formally introduced.
- This AI model is comprised of three different "sizes," including Gemini Nano, Pro, and Ultra.
- Gemini Nano is designed for mobile devices, while Gemini Pro is being integrated into Google Bard.
- Gemini Ultra is set to arrive sometime next year, and will be part of the new "Bard Advanced experience."
The wait is finally over, as Google has officially unveiled its true ChatGPT competitor with Google Gemini. Announced at Google I/O 2023, Gemini is the company's "next-generation foundation model," which is comprised of three different models: Ultra, Pro, and Nano.
While it might seem as though Google moved swiftly to bring its new AI model to the public, this couldn't be further from the truth. With the help of Google DeepMind, Gemini has been in the works since 2012 before ramping up development in 2015 with AI accelerators.
Unlike Bard, which has underwhelmed when compared to the likes of ChatGPT and Claude, Gemini brings true competition. Of the three different models, Google provides a broad overview of what each model can deliver:
- Gemini Nano: Most efficient model for on-device tasks.
- Gemini Pro: Best model for scaling across a wide range of tasks.
- Gemini Ultra: Largest and most capable model for highly complex tasks.
Breaking things down a bit, Gemini Nano is what we'll see come to the best Android phones. Not only is Gemini Nano being opened to developers via AICore, but Google already confirmed that it will be coming to the Pixel 8 Pro.
However, don't expect a dedicated app for Gemini access, as Google states that it's "powering new features like Summarize in the Recorder app and rolling out Smart Reply in Gboard, starting with WhatsApp." But, thanks to the ability for developers to implement and "build with Gemini Nano" through AICore, we're hoping it won't be long before Nano comes to more apps.
Moving up a step, Gemini Pro will be made available to "developers and enterprise customers" starting on December 13. It will be accessible through the new Gemini API found in either Google AI Studio or Google Cloud Vertex AI.
Thankfully, we won't have to wait for the opportunity to play around with Gemini Pro. Starting today, Gemini Pro has been implemented into Google Bard "for text-based prompts, with support for other modalities coming soon." Currently, Bard with Gemini Pro is only available in 170 countries, but Google plans to expand to more regions and with more languages "in the near future."
A lot of focus has been put on Gemini Pro, as Google claims it's the "best model for scaling across a wide range of tasks." One of the studies provided by Google states that Gemini Pro was able to outperform GPT-3.5 "in six out of eight benchmarks." While impressive, we also have to wonder about whether it will be improved upon to surpass GPT-3.5 or if the primary focus will be on Nano and Ultra.
Speaking of which, Gemini Ultra is arguably the most exciting model of the bunch, as it's the "most capable model." However, we will have to wait a bit longer before getting our hands on Ultra, as Google is "currently completing extensive trust and safety checks."
Gemini Ultra will first come to "select customers, developers, partners and safety and responsibility experts for early experimentation and feedback before rolling it out to developers and enterprise customers early next year." But there's something else to be excited about, as Ultra is expected to be implemented into a new "Bard Advanced experience," which is expected to arrive early next year.
Part of what makes Gemini Ultra so intriguing is how powerful it seems to be. Google states that Ultra is "the first model to outperform human experts on MMLU (massive multitask language understanding), which uses a combination of 57 subjects such as math, physics, history, law, medicine and ethics for testing both world knowledge and problem-solving abilities."
How did it do this? By receiving a score of 90%, which is higher than the 86.4% calculated score achieved by OpenAI's GPT-4 model. Gemini outperformed GPT-4 in every category except for the "HellaSwag" benchmark, which is described as "commonsense reasoning for everyday tasks."
If it weren't evident enough by now, Google Bard was merely a stopgap as the company continued to develop Gemini to the point of providing a public release. But Bard is also serving as the access point between us and Gemini, as you can start using Bard with Gemini Pro today.
To help provide a bit of fun and show off what Bard with Gemini Pro is capable of, Google partnered with YouTuber Mark Rober. In the video, Rober uses Bard to come up with some video ideas, resulting in Rober working to create "The World's Most Accurate Paper Airplane."
From there, Bard is used to generate "story structures," which Rober then proceeds to follow. In each of the sections, Bard isn't solely relying on text prompts, as Rober takes pictures of the different paper airplane designs and has Bard analyze them to determine which is the best.
After discovering the most accurate paper airplane design, Bard is then used to "critique" the design. Bard spits out its findings while also offering a few suggestions, such as "the wings are too small" and "the paper is too light." From there, Rober takes Bard's advice, showcasing the various testing methods that were taken before the big finale.
While the teaser video is great, it's not exactly the perfect example of what we're able to try via Bard with Gemini Pro today. However, it gives us a glimpse into the bigger picture of what will hopefully be possible once Bard Advanced with Gemini Ultra is made available.
Nevertheless, it's refreshing to see Gemini make its official debut before we flip the calendar into 2024. This announcement has the potential to be the biggest thing to happen to Google in the past ten years. Now, we just have to hope that Bard with Gemini can live up to the hype.
