The Dawn of Gemini Pro
So, Google decided to grace us with Gemini Pro, its latest brainchild in the GenAI model family, through its Vertex AI platform. It's like the younger sibling of the more capable Gemini Ultra, but don't let that fool you. This "lightweight" version is now strutting its stuff in public preview, offering features like chat functionality and filtering, and it supports a whopping 38 languages and regions. Thomas Kurian, the big boss at Google Cloud, proudly proclaims that Gemini Pro is a state-of-the-art, natively multimodal model with "sophisticated reasoning advanced coding skills". In other words, it’s smart – real smart. Developers, start your engines! You can now build your own applications with this shiny new tool.
Gemini Pro's Party Tricks
But wait, there's more! Gemini Pro isn't just about text; it's got an eye for images too. The new kid on the block, Gemini Pro Vision, is stepping up the game by processing text and imagery, including photos and videos. Imagine that – a model that can understand your holiday snaps and cat videos. Gemini's image processing feature addresses past criticisms about its inability to handle images, despite being labeled "multimodal". So, now users can test the model's image comprehension skills for themselves. Plus, developers can customize Gemini Pro for specific contexts using tools available for other Vertex-hosted models like Google’s PaLM 2. It's like a digital chameleon, adapting to your needs.
Fact-Checking and Control
Google also introduces citation checking for Gemini Pro, which is like having a stern librarian making sure your facts are straight. This feature highlights the sources of information Gemini Pro used to arrive at a response. Kurian explains that "grounding" allows comparison of Gemini's answers with data within a company’s systems or web sources. It's like having a fact-checker in your pocket, ensuring the model's answers are not just clever, but also correct.
Attractive Pricing: The Sweetener
Amid concerns that Gemini Pro might not be the strongest model out there, Google is playing the pricing card. They're offering input for Gemini Pro on Vertex AI at $0.00025 per character and output at $0.00005 per character. That’s a price slash from its predecessor’s rates. And as a cherry on top, it's free for a limited time. Google seems to be saying, "Come and get it while it's hot (and cheap)!".
Future Plans and Rivalry
Looking ahead, Google plans to enhance Vertex AI with features like custom-built conversational voice and chat agents. Gemini Pro is expected to power dynamic interactions supporting advanced reasoning and will play a role in search summarization, recommendation, and answer generation. The goal? To make developers turn their backs on rival platforms. As for when these features will drop, Kurian hints at "very early" in 2024.
Google's Answer to Model Evaluation: Auto SxS
And just when you thought that was all, Google introduces Automatic Side by Side (Auto SxS) in Vertex. This is Google’s retort to AWS’ Model Evaluation, allowing developers to evaluate models in an automated fashion. It's touted as faster and more cost-efficient than manual evaluation. Plus, Google’s indemnification policy now covers outputs from models like PaLM 2 and Imagen, offering legal defense in IP disputes. This is Google’s way of saying, “Don’t worry, we got your back” to corporate customers wary of IP issues.
And there you have it, folks! Google's Gemini Pro – a mix of brainy features, attractive pricing, and future promises. Will it be enough to woo developers and outshine its rivals? Only time will tell. But for now, Google seems to be putting its best foot forward, or should I say, its best algorithm forward.