Decrypt's Art, Fashion, and Entertainment Hub. Introducing SCENE In line with its promise, Google has opened up its highly advanced AI model, Gemini 1. 5 Pro, to the public following a successful beta release for developers last month. Compared to previous AI models, Google's Gemini 1. 5 Pro is capable of handling more complex tasks, including the analysis of extensive text libraries, feature-length Hollywood movies, and nearly a full day's worth of audio data. It surpasses OpenAI's GPT-4o in terms of data capacity by a factor of 20, and Anthropic's Claude 3. 5 Sonnet by nearly 10 times. The objective is to provide AI developers with faster and more cost-effective tools, enabling new use cases along with increased production robustness and reliability, as stated in Google's announcement. Image: Google AD AD Back in May, Google initially introduced the model, showcasing videos that demonstrated how a select group of beta testers were able to harness its capabilities. For instance, machine-learning engineer Lukas Atkins fed the model the entire Python library and asked it questions to help him solve a problem. In the video, he expressed satisfaction with the model's performance, highlighting its ability to find specific code comments and fulfill specific user requests. Another beta tester filmed his entire bookshelf, and Gemini created a comprehensive database of all the books he owned—a feat that is nearly impossible to achieve with conventional AI chatbots. Gemma 2 Emerges as a Dominant Open Source Solution Google has also made significant strides within the open source community.

The company recently released Gemma 2 27B, an open source large language model that quickly claimed the title of the best quality responses among open source models, according to the LLM Arena ranking. Google asserts that Gemma 2 delivers top-notch performance, operates at exceptional speeds across different hardware, and seamlessly integrates with other AI tools. It is positioned to compete with models twice its size, according to the company. AD AD Image: Google The license for Gemma 2 allows for free access and redistribution, although it differs from traditional open-source licenses such as MIT or Apache. The purpose of this model is to facilitate more accessible and cost-friendly AI deployments, available in both its 27B and smaller 9B versions. This is significant for both individual users and enterprises because, unlike closed models, a powerful open model like Gemma can be extensively customized. Users have the ability to fine-tune the model to excel in specific tasks, ensuring data privacy by running these models locally. For instance, Microsoft's small language model, Phi-3, has been fine-tuned specifically for mathematical problems, outperforming larger models like Llama-3 and even Gemma 2 itself in that domain. Image: Microsoft Gemma 2 is now accessible through Google AI Studio, with model weights available for download from platforms like Kaggle and Hugging Face Models. Developers can test it using the powerful Gemini 1. 5 Pro on Vertex AI.